physics 426 notes: plasma astrophysicskestrel.nmt.edu/~lyoung/426/root.pdfphysics 426 notes: plasma...
TRANSCRIPT
PHYSICS 426 NOTES: PLASMA ASTROPHYSICS
Original version (2008): Jean Eilek
Physics Department, New Mexico Tech
Socorro, NM 87801, U.S.A.
minor revisions (2009-2013): Lisa Young
These notes continue what we started in Physics 425.
Our goal is still to explore the “physics of astro-
physics”. This term we’ll develop the radiation-
physics tools we’ll need, and focus more on galax-
ies, their interstellar medium, and high-energy astro-
physics.
Contents
1 Galaxies, normal and otherwise 1
1.1 Normal galaxies: spirals . . . . . . . 1
1.2 Normal galaxies: ellipticals . . . . . . 2
1.3 Query: why are they different? . . . . 3
1.4 Interlude: stellar dynamics . . . . . . 3
1.5 The not-so-normal ones: active galaxies 4
1.6 The beast in the core . . . . . . . . . 5
1.6.1 Techniques: normal galaxies . 5
1.6.2 Techniques: extend to AGN . 5
1.6.3 The galactic center: stellar orbits 6
2 The Interstellar Medium 8
2.1 The diffuse ISM in our galaxy . . . . 8
2.1.1 How we oberve the ISM . . . 8
2.1.2 A multi-phase equilibrium? . 8
2.1.3 Other components of the ISM 9
2.2 “Star stuff” in our galaxy . . . . . . . 9
2.3 Galactic ecology . . . . . . . . . . . 10
2.4 The ISM in Ellipticals . . . . . . . . . 10
2.4.1 Everything that isn’t the hot
phase . . . . . . . . . . . . . 10
2.4.2 The hot phase – the x-ray loud
gas . . . . . . . . . . . . . . 11
3 Some Radiation Basics 12
3.1 Radiation: some important definitions 12
3.2 Thermal equilibrium: an ideal gas . . 12
3.3 Thermal equilibrium: radiation . . . . 13
3.4 Radiative transfer . . . . . . . . . . . 14
3.4.1 More definitions . . . . . . . 14
3.4.2 Transfer analysis . . . . . . . 15
3.4.3 Optically thick and thin limits 15
3.5 Appendix I: some examples with in-
tensity . . . . . . . . . . . . . . . . . 16
3.5.1 Isotropic radiation field . . . . 16
3.5.2 Intensity is constant along a ray 16
3.5.3 Flux from a sphere . . . . . . 16
3.6 Appendix II: more on pressure . . . . 17
3.6.1 Ideal gases: the pressure integral 17
3.6.2 Radiation pressure . . . . . . 17
4 Bremsstrahlung radiation 19
4.1 Some basic tools . . . . . . . . . . . 19
4.1.1 Power; Larmor formula . . . . 19
4.1.2 Spectrum: Fourier analysis . . 19
4.2 Bremsstrahlung I: single particle . . . 20
4.3 Bremsstrahlung II: from a plasma . . 21
5 Thermal state of the ISM 24
5.1 Heating and cooling: general consid-
erations . . . . . . . . . . . . . . . . 24
5.1.1 Cooling . . . . . . . . . . . . 24
5.1.2 Heating . . . . . . . . . . . . 25
5.2 HII regions . . . . . . . . . . . . . . 25
5.2.1 Ionization structure . . . . . . 26
i
ii Physics 426 Notes Spring 2016
5.2.2 Energy balance and temperature 27
5.3 The diffuse ISM: multiphase equilibrium 28
5.3.1 Cooling: what dominates here? 28
5.3.2 Heating: by cosmic rays? . . . 29
5.3.3 Thermal balance and multi-
phase
equilibrium . . . . . . . . . . 29
5.3.4 Is the thermal balance solution
stable? . . . . . . . . . . . . 30
6 Dynamics of the ISM: energetics & shocks 32
6.1 Fluids: energetics . . . . . . . . . . . 32
6.2 Supersonic flow and shock fronts . . . 33
6.2.1 Adiabatic shocks . . . . . . . 33
6.2.2 Isothermal shocks . . . . . . . 34
6.2.3 Magnetized shocks . . . . . . 34
6.2.4 Oblique shocks . . . . . . . . 34
7 Stellar Winds & Supernovae Remnants 36
7.1 Stellar winds and the surrounding ISM 36
7.1.1 The basic solution . . . . . . 36
7.1.2 The outer shock . . . . . . . . 36
7.1.3 What about the inner shock? . 37
7.2 Supernova remnants . . . . . . . . . . 38
7.2.1 Early: energy conserving (Se-
dov) phase. . . . . . . . . . . 38
7.2.2 Late: momentum conserving
(snowplow) phase. . . . . . . 39
7.3 Plerions, a.k.a. pulsar wind nebulae . 39
8 Relativistic particles in astrophysics 41
8.1 Recap: basics for relativistic particles 41
8.2 Quick overview of the observations . . 41
8.3 Cosmic rays in the galactic setting . . 42
8.4 Particle acceleration, overview . . . . 42
8.5 Particle acceleration, first stage mech-
anisms . . . . . . . . . . . . . . . . . 42
8.5.1 Magnetic reconnection . . . . 43
8.5.2 Unipolar dynamos . . . . . . 43
8.6 Particle acceleration, second stage
mechanisms . . . . . . . . . . . . . . 44
8.6.1 Fermi acceleration . . . . . . 44
8.6.2 Plasma turbulence: Alfven
waves . . . . . . . . . . . . . 45
8.6.3 Turbulent shock acceleration . 45
8.6.4 Energy limits for Alfven ac-
celeration . . . . . . . . . . . 46
9 Synchrotron radiation 47
9.1 Total power . . . . . . . . . . . . . . 47
9.2 Single particle spectrum . . . . . . . 47
9.3 Spectrum from a distribution of parti-
cle
energies . . . . . . . . . . . . . . . . 48
9.4 Polarization . . . . . . . . . . . . . . 49
9.5 Synchrotron self-absorption . . . . . 49
9.6 Total synchrotron spectrum . . . . . . 50
10 Pair plasmas in Astrophysics 52
10.1 Pair annihilation . . . . . . . . . . . . 52
10.2 Pair creation . . . . . . . . . . . . . . 52
10.2.1 Two-photon pair production . 52
10.2.2 Pair production in pion decay 53
10.3 Magnetic pair production . . . . . . . 53
11 (Inverse) Compton scattering 55
11.1 Basic Tools . . . . . . . . . . . . . . 55
11.1.1 One event seen in the ERF . . 55
11.1.2 Cross sections . . . . . . . . 55
11.1.3 Remember your relativity . . 55
11.2 Scattering as seen in the lab . . . . . . 56
11.2.1 Single particle radiation . . . 56
11.2.2 Single particle spectrum . . . 57
11.3 Composite spectra . . . . . . . . . . . 57
11.3.1 Nonrelativistic electrons . . . 57
11.3.2 Relativistic electrons, single
scattering . . . . . . . . . . . 57
11.3.3 Scattering from power-law
electrons . . . . . . . . . . . 57
12 Pulsars: overview and some physics 59
12.1 The basic picture . . . . . . . . . . . 59
12.1.1 The cartoon . . . . . . . . . . 59
12.1.2 And some details . . . . . . . 59
12.2 Spin a magnetic field . . . . . . . . . 60
Physics 426 Notes Spring 2016 iii
12.2.1 Star in vacuum . . . . . . . . 60
12.2.2 Filled magnetosphere . . . . . 61
12.3 Radio emission and the pair cascade . 61
12.4 High altitudes and currents . . . . . . 62
12.4.1 High energy emission . . . . 62
12.4.2 The pulsar circuit? . . . . . . 62
12.5 Winds and nebulae . . . . . . . . . . 62
12.5.1 Pulsar winds . . . . . . . . . 63
12.5.2 Pulsar wind nebulae . . . . . 63
12.6 Magnetars and Anomalous pulsars . . 63
13 Radio jets and radio galaxies 65
13.1 Jets: the observational constraints . . 65
13.2 Some useful relativity . . . . . . . . . 66
13.2.1 Superluminal motion . . . . . 66
13.2.2 Doppler beaming . . . . . . . 66
13.3 Some useful physics . . . . . . . . . 66
13.3.1 Collimation . . . . . . . . . . 66
13.3.2 Jet transport . . . . . . . . . . 67
13.4 Larger Scales: the Radio Galaxy . . . 67
13.4.1 Classical Double radio galax-
ies (FR II’s) . . . . . . . . . . 67
13.4.2 Tailed radio galaxies (FR I’s) . 68
13.5 Unresolved issues . . . . . . . . . . . 69
13.5.1 What is the life cycle of a RG? 69
13.5.2 How does a jet affect its envi-
ronment? . . . . . . . . . . . 69
13.6 How are jets made? . . . . . . . . . . 69
13.6.1 Wind (fluid-based) models . . 70
13.6.2 MHD models . . . . . . . . . 70
13.6.3 Duty cycles? . . . . . . . . . 70
14 Quasars and Active Galactic Nuclei 72
14.1 Basic properties: observations . . . . 72
14.1.1 Spectral lines . . . . . . . . . 72
14.1.2 Continuum emission . . . . . 72
14.2 The AGN zoo . . . . . . . . . . . . . 73
14.2.1 The radio-quiet ones . . . . . 73
14.2.2 The radio-loud ones . . . . . 73
14.2.3 Blazars and friends . . . . . . 73
14.2.4 Parent galaxies . . . . . . . . 73
14.3 The usual model: a massive BH . . . 74
14.3.1 Zoom in: the central kpc and
within . . . . . . . . . . . . . 74
14.3.2 Zoom in further: the central pc
and within . . . . . . . . . . 74
14.3.3 Why radio-loud vs. radio-quiet? 74
14.3.4 Why are only some galaxies
“active”? . . . . . . . . . . . 74
14.4 Unification Models . . . . . . . . . . 74
14.4.1 Relativistic beaming . . . . . 75
14.4.2 Obscuration and tori . . . . . 75
14.5 AGN demographics . . . . . . . . . . 75
14.5.1 Was there a “quasar era?” . . 75
14.5.2 What about galaxy formation? 76
14.6 Ending with questions . . . . . . . . . 76
14.7 Appendix: a little practical cosmology 77
14.7.1 Just what is the redshift? . . . 77
14.7.2 The Hubble diagram . . . . . 77
14.7.3 The lookback time . . . . . . 77
Physics 426 Notes Spring 2016 1
1 Galaxies, normal and otherwise
To start, let’s recall the large-scale structure of galax-
ies. We are going to focus on bright galaxies, spirals
and ellipticals. A lot is known about these objects; in
addition to being pretty, they are bright and extended,
thus easy to study. We should remember, however,
that most galaxies in the universe are neither S nor E.
By counting galaxies we can determine the luminosity
function, that is the number of galaxies at luminosity
L (per volume usually). This is called the Schechter
function, and has the approximate form,
Φ(L) = Φo
(
L
L∗
)−α
e−L/L∗ (1.1)
Here, Φo is a normalizing constant (which depends
on the local environment: for instance Φo is much
larger for a rich cluster than for the field). The slope
α ∼ 1.25; and L∗ ∼ 7 × 1010L⊙ is comparable to
the characteristic luminosity of bright galaxies (values
quoted by Elmegreen, for the V band). We know that
the galaxy luminosity connects directly to the galaxy’s
mass: M/L ∼ 30 − 100 (in solar units; depending
somewhat on the galaxy type, and with some scatter).
Thus the Schecter function also measures the mass dis-
tribution of galaxies.
Important point: this is a composite luminosity func-
tion, determined by adding all galaxy types. Most S
and E galaxies sit within a factor of a few of L∗. But
the LF has been measured over a range ∼ 10 magni-
tudes, that is a factor ∼ 104 in luminosity, and it keeps
rising to smaller luminosities. It follows that most of
the galaxies in the universe are neither spirals nor el-
lipticals. Most galaxies are either Irregulars (small,
patchy structure, rotation supported) or dwarf Ellip-
ticals (small, featureless, probably not rotation sup-
ported). Check Figure 23.34 of Carroll & Ostlie for
a recent break-down of the total LF by galaxy type. In
this course we will focus on the big galaxies, which are
better understood; but don’t forget the little guys.
1.1 Normal galaxies: spirals
Our galaxy is a medium-size spiral, and we can use it
to study a “typical” spiral. As with all spirals, its most
notable feature is its disk, which contains both stars
and gas. The surface mass density of the stellar disk
is exponential, Σ(R) ∝ e−R/H , with H ∼ few kpc
(that number is typical of other nearby spirals). Thus
the density of stars dies after several H distances. The
gas (HI and molecular gas), however, extends much
further; in big spirals gas can be traced to 40-50 kpc,
and it has been traced to at least ∼ 20 kpc in our galaxy.
The surface density of the gas falls, in some galaxies
exponentially and in others more irregularly; its radial
scale length is generally larger than that of the stars.
What is the structure of the gravitating matter?
The disk is supported vertically by the random motions
(“heat”) of the stars and gas. Think about individual
stars: each one moves up and down, through the galac-
tic plane, in approximate harmonic motion. The ver-
tical extent of its motion depends on the local gravity.
Or, think about the stars as a “gas”,1 with random mo-
tions at speed σ. We can describe the ensemble of stars
in a given volume by a “pressure” (p ↔ nmσ2 = ρσ2,
for number density n and stellar mass m). We then
expect the vertical disk structure to be described by
hydrostatic equilibrium. This last can be written as a
vector equation, or alternatively we can isolate its z(“vertical”) component:
∇p = ρg ;dp
dz= ρgz (1.2)
The typical scale height of the disk, locally, is H ∼200− 300 pc (varying somewhat for different types of
stars, or different phases of the ISM). The surface mass
density in the disk can be found in two ways. One, we
can measure the local density of stars and gas directly,
to get a density of luminous mass. Two, we can use the
vertical support condition (1.2) to find the local gravity,
and from this (remembering Poisson’s law for gravity,
∇2ΦG = −∇ · g = 4πGρ), we can find the density
of gravitating mass. This latter approach gives Σ ∼75M⊙/pc2, while the former (counting stars) gives a
value that is only about half of this. Thus, about half
of the gravitating mass in the local disk is dark: it does
not emit any radiation that we have been able to detect.
In addition, the galaxy has more dark matter on larger
scales. Consider our galaxy: the disk is supported
“horizontally” by its rotation; our local rotation speed
∼ 220 km/s, which gives a rotation period at the sun’s
distance (8.5 kpc) of ∼ 230 Myr. The stellar orbits
are nearly circular, and the galaxy is close to axisym-
metric (or truly, we are assuming its mass distribution
is spherically symmetric!), so we can use the simplest
form of Kepler’s law:
v2rot =GM(r)
r(1.3)
1If you don’t like this idea, look at §1.4 for some discussion.
2 Physics 426 Notes Spring 2016
where M(r) is the mass within r. This provides a sim-
ple way to estimate the gravitating mass of the galaxy.
Turning to nearby external spiral galaxies, we can use
the rotation curve to find the gravitating mass. Rota-
tion curves in big spirals can be traced out to several
tens of kpc using HI, and their rotation velocity vrotstays constant out that far. Thus, the gravitating mass
M(< r) ∝ r: the mass of the galaxy keeps rising as
we go to larger distances. This is not what happens
to the total luminous mass (in stars plus gas), how-
ever: the integral of an exponential converges to a fi-
nite value. Thus, the ratio of total mass to luminous
mass increases as we go to larger scales. By the outer
∼ 40− 50 kpc in big systems, the ratio of (gravitating
mass)/(luminous mass) ∼ O(10); there is something
like ten times more mass in the system than we can
see.
A note on spiral arms...they are of course the most
striking, defining features of spiral galaxies. They
are not, however, fundametal to the galaxy’s structure.
Rather they are waves or perturbations in the self-
gravitating disk of the galaxy. Some authors like to
work with linear density waves – i.e. small-amplitude
waves with a spiral shape. Other authors like to work
with global perturbations of the galactic disk – which
have a generally spiral shape but need not be linear.
Still other authors consider local perturbations – in
which a local overdensity, say, is enhanced (due to its
own self-gravity) and sheared into a spiral fragment
(due to the differential rotation of the disk). Probably
each approach describes some fraction of spiral galax-
ies.
And a note on the dark matter ... we have no direct ob-
servation of the spatial distribution of the dark matter.
It seems likely, however, that the dark haloes of spiral
galaxies are spheroidal – i.e. supported by their inter-
nal, random motions (“heat”), just as elliptical galaxies
are (as we discuss in the next section). This idea comes
from numerical simulations of structure formation, as
well as the fact that the dark matter (by assumption!)
does not “cool”, so can’t dissipate its internal energy –
thus it probably has not been able to flatten into a disk.
1.2 Normal galaxies: ellipticals
Elliptical galaxies are quite different. They are smooth
and featureless structures, showing a core (of roughly
constant density) and an outer envelope (or declining
density). Their surface brightness follows the heuristic
De Vaucoulers law: Σ(r) = Σoe−(r/ro)1/4 (where ro is
a constant, a length scale). In a three-dimensional sys-
tem such as this, the surface brightness is a projection
of the underlying 3D spatial density:
Σ(R) = 2
∫ ∞
Rn(r)
rdr√r2 −R2
(1.4)
The stellar density is decently well fit by n(r) =no/[r(r + a)2], or by similar (analytic) forms which
have a characteristic “core” radius, a ∼ 1− 2 kpc. 2
We might think they are very simple...but that’s not the
case. One hint comes from rotation: elliptical galax-
ies are not rotation supported. In a large elliptical
the rotation speed is much smaller than the dispersion:
vrot <∼ 0.1σ typically. (For smaller galaxies, vrot is a
larger fraction of σ). In addition, E galaxies are truly
three-dimensional. To see this, consider shape of the
surface brightness isophotes. They are, of course, el-
liptical (and close to circular in some E galaxies). But,
the direction of their major axis rotates going out from
the galactic center. This isophote twist is a clear sign of
a trixial system. Both of these facts tell us that the stel-
lar orbits must be complex, randomly oriented, and not
necessarily closed (that is a star can wander through
much of the volume of the galaxy over it’s lifetime).
Despite the complexity of the orbits, we can find a sim-
ple model of the structure of the galaxy. Assume spher-
ical symmetry to simplify, and following the arguments
above write down a “hydrostatic balance” equation for
the stars:
d(ρσ2)
dr= ρ
GM(r)
r2;
dM
dr= 4πρr2 (1.5)
Now do some algebra, and combine these into one
second-order ODE for ρ(r). The resulting equation
has two solutions. One is analytic, ρ ∝ 1/r2. This
solution diverges in the center (that’s not good) and its
mass, M(r) ∝∫
ρr2dr, diverges at infinity (that’s also
not good). The other must be found numerically, but
turns out to have a finite central density, and a charac-
teristic core radius; at large radii it approaches the first,
analytic, solution. And: you have no doubt recognized
this solution, from last term. It is the self-gravitating
isothermal sphere, an important solution in several dif-
ferent astrophysical applications.
2This is the “NFW” (Navarro, Frenk & White) profile; it turns
up commonly in numerical, N-body simulations of collapsing,
self-gravitating galaxies; and is a reasonable match to the profiles
of real galaxies.
Physics 426 Notes Spring 2016 3
The divergence at infinity cannot be fixed analytically
in this approach. The most attractive solution that I
know of, comes from considering the internal veloc-
ity distribution of the stars; those at the highest ve-
locities will exceed the escape velocity of the galaxy.
King took this into account in numerical models of
the isothermal sphere, and found solutions which are
nicely well-behaved at infinity. He also gives an ap-
proximate expression for the density of the resulting
system:
ρKing(r) =ρoa
3
(r2 + a2)3/2(1.6)
What about the mass of an elliptical? Is dark matter
important? Due to the complexity and variety of the
shapes of allowed stellar orbits, we can’t easily use
Kepler-type arguments, as we could for spirals. We can
of course use approximate arguments, such as the virial
theorem, to estimate Mgrav from the stellar velocity
dispersion: GMgrav ≃ σ2r. Quantitatively, however,
to get accuracies at the tens of percent level is not as
easy as it is for spirals. Several approaches can be used:
• One, the strongest result in my opinion, uses the
small (but non-trivial) minority of the population
which have flattened, extended HI disks. These disks
rotate, in simple Keplerian motion, which allows us to
find the gravitating mass directly. The result: M/Lfor big E’s is similar to that for big S’s, (∼ 10 − 30typically), and also tends to increase with radius.
• Another approach uses the hot, X-ray loud gas found
in every big elliptical. This gas sits in approximate
hydrostatic equilibrium in the potential well of the
galaxy; from its spatial distribution we can estimate
the gravitating mass. I defer details here until the next
chapter; the results for M/L are generally consistent
with those from flattened HI disks.
• Careful interpretation of the stellar kinematics in-
volves making a model of the gravitational potential,
doing numerical integrations to find out what the al-
lowed stellar orbits look like, and verifying that if one
populated those orbits with stars the resulting object
would look like a real galaxy and would reproduce the
potential assumed at the beginning. At present it is not
very common to be able to do this in the far outer re-
gions of elliptical galaxies. In the inner, bright portions
of the galaxy the mass is dominated by stars rather than
by dark matter.
1.3 Query: why are they different?
Although this isn’t a course in galactic structure, it
seems appropriate to ask why there are two character-
istic types of big galaxies – one flat and supported by
rotation, the other round and supported by random mo-
tions. Two possibilities have been discussed for ages:
• We might think that galaxies form in isolation: by
gravitational collapse of their dark matter, and subse-
quent dissipational collapse of the baryonic gas (nor-
mal ISM!) within the dark halo. (An important point
here is that normal, baryonic material can radiate away
its interal energy; so it can cool, and collapse in a grav-
itational field. Dark matter, being non-baryonic, can-
not). If this is the case, then the E/S difference might
be due to how early in the process stars formed. Early
star formation might lead to an E; later star formation,
after the ISM has collapsed into a disk, could lead to
an S.
• However there’s another possibility: galaxies might
influence each other during their formation process.
Two facts are germane here. First, we know that E’s
are commonly found in regions of high galaxy den-
sity – rich clusters of galaxies. Second, simulations
show that if two spiral galaxies collide, closely enough
for their stars to remain bound, the energy of the col-
lision will heat the stars and “puff up” the galaxy –
i.e. making something that looks like an elliptical. So
.... it may well be that E’s are more common in clus-
ters, because conditions (at least early on) were right
for protogalaxy-protogalaxy collisions.
Which of these is right? I suspect the answer lies some-
where inbetween – this is still an active area of current
research.
1.4 Interlude: stellar dynamics
How can we justify talking of a stellar “pressure”?
Here’s a quick overview of the argument, following
Binney & Tremaine. Consider a distribution function
of stellar velocities, f(x,v) (the number of stars “at x
and v”). If stars are conserved, and there are no colli-
sions (in which the position and/or the velocity of the
star changes instantaneously, by a finite amount), then
the evolution of f is goverened by the motion of indi-
vidual stars through (x,v) phase space:
∂f
∂t+ v · ∇xf + a · ∇vf = 0 (1.7)
4 Physics 426 Notes Spring 2016
(Think: does this make sense? We’re just counting
stars). Here, ∇x is the usual spatial gradient, ∇v is
the gradient in phase space with respect to the veloc-
ity coordinates, v is the velocity and a ≡ dv/dt is the
acceleration. Here, we specify a = g = −∇Φg, i.e.
the gravitational acceleration. Now, we can derive a
couple of useful results.
First, integrate (1.7) over all velocities. We get (spec-
ify to Cartesian coordinates to make the algebra more
transparent):∫
∂f
∂td3v+
∫
vi∂f
∂xid3v+ gi
∫
∂f
∂vid3v = 0 (1.8)
But now, the last term goes to zero (why? The i compo-
nent of the integrand becomes (∂f/∂vi)dvi, a perfect
differential; it integrates to zero if f has “reasonable”
boundary conditions). Also, in the second term we can
take the d/dxi outside the integral (remember that x
and v are independent coordinates). We can define the
stellar density and mean velocity by
n =
∫
fd3v ; 〈v〉 = 1
n
∫
fvd3v (1.9)
Finally, (1.8) becomes
∂n
∂t+
∂
∂xi(n〈vi〉) =
∂n
∂t+∇ · (n〈v〉) = 0 (1.10)
(here and after I’m using ∇ to mean the usual ∇x).
But this is simple: it is just the continuity equation for
stars.
Now, multiply (1.7) by v before integrating over ve-
locity space. The algebra is longer, but the approach
(and mathematical tricks) are the same. One version of
the result is
n∂〈vj〉∂t
− 〈vj〉∂
∂xi(n〈vi〉) +
∂
∂xi(n〈vivj〉) = ngj
(1.11)
Now, velocity v is partly due to streaming (all stars
share the same mean velocity) and partly due to ran-
dom motions about this mean. Thus, the mean value
〈vivj〉 can be split into a part that describes the stream-
ing motion (〈vi〉〈vj〉) and a part that describes the in-
ternal velocity dispersion,
σ2ij = 〈vivj〉 − 〈vi〉〈vj〉 (1.12)
Using this, and letting v now represent the mean
streaming velocity, we can write (1.11) as
n∂〈vj〉∂t
+ n〈vi〉∂〈vj〉∂xi
= ngi −∂
∂xi
(
nσ2ij
)
(1.13)
or – if we note that nmσ2ij is equivalent to a pressure,
p – this becomes
n∂v
∂t+ nv · ∇v = ng− 1
m∇p (1.14)
This is the general dynamical equation for collision-
less stellar systems. In a steady state, with no bulk
flows, this reduces to the usual hydrostatic equilibrium:
∇p = ρg (being slightly cavalier about the pressure,
which is OK for our purposes here).
1.5 The not-so-normal ones: active galaxies
About one per cent of the bright galaxy population
is “active”. That is, they contain small, highly ener-
getic, non-stellar events in their nuclei. (The fraction
is less common in smaller galaxies; and more common
in bright galaxies at high redshift). Some active galax-
ies are found optically, others are found in radio. We
will return to this topic later in the course. For now,
here, I just note the classes and general properties, with
an eye to their historical discovery.
Seyfert galaxies are spirals with very bright nuclei.
These nuclei are most easily detected by their strong,
optical emission lines; they also have nonthermal con-
tinuum emission. They also have small ( <∼ kpc), faint
radio sources in their cores. The most important point
here is that this phenomenon cannot be explained sim-
ply by stars: some non-stellar event is taking place in
these nuclei.
Radio galaxies are ellipticals which have double-
lobed radio sources, fed by jets emanating from galac-
tic core. They also have compact, non-stellar nuclei,
but on the weak side compared to bright Seyferts.
Quasars were originally called “quasi-stellar objects”.
They are bright, compact (inferred from from variabil-
ity), have very strong emission lines (this is how they
were first found), and a strong nonthermal continuum.
Some (maybe 10%) are “radio-strong”, with a radio-
loud core and extended double-lobed radio structure.
Some of these (“blazars”; maybe 10% again?) are vi-
olently variable, showing a large ∆L/L on short time
scales. These generally also contain radio jets with su-
perluminal motion. Initially no galaxy could be seen
around these very bright, small sources – hence the
name QSO. With dedication and better technology we
now can image the underlying galaxies. This area is
still under discussion, but it looks as if quasars show
the same host-galaxy split as to Seyferts and AGN:
Physics 426 Notes Spring 2016 5
radio-loud from E’s and radio-weak from S’s. We now
have found quasars out to z ∼ 5, and probably higher
by the time you read these notes.
What all of these objects have in common is an Active
Galactic Nucleus (AGN). An AGN is characterized by
a high luminosity (which can be comparable to the lu-
minosity of the entire host galaxy); a small intrinsic
size (inferred from variability: <∼ light-day or light-
month); a non-stellar spectrum (that is, from diffuse
gas, often including nonthermal particles and B field);
and a preferred axis (as shown by radio jets – showing
us net angular momentum of the core?). The general
model is accretion of matter onto massive black hole
in the nucleus of the galaxy.
1.6 The beast in the core
A striking recent result is that every galaxy appears to
have a massive dark object in its core. At this point
we have no definite proof that these massive things
are black holes; that would require resolving the event
horizon and finding some definitive signature, for in-
stance emission lines red/blue shifted due to the orbital
speed of gas in the last stable orbit. What we can do,
however, is use gravity to detect a massive dark object
(MDO): that is, a total gravitating mass much larger
than what can be accounted for by stars in the region.
1.6.1 Techniques: normal galaxies
There are two important detection techniques here, for
galaxies without strong AGN. In what follows I use the
term “bulges” to mean either elliptical galaxies, or the
bulges of early-type spirals.
•gas disks Some bulges contain gas disks in (apparent)
Kepler rotation around a central MDO. A small num-
ber of these, such as M87, have been resolved fairly
close to the MDO. A few other galaxies have inner gas
disks with maser spots which are easy to resolve with
the VLBA and which can also give the gas velocities.
A larger number of galaxies have gas disks which can
be resolved with HST (albeit not so close to the MDO).
Emission line velocities from these disks, again assum-
ing Keplerian rotation, can be used to find the central
mass.
•stellar cusps Most E’s and spiral bulges, however, do
not have nice gas disks in their cores. For these, we
need to use the fact that a central point mass affects the
distribution of nearby stars. A central star cluster with
no MDO is well described by a self-gravitating isother-
mal sphere – which you remember from earlier in the
course. The stellar density is (approximately) constant
in the inner region of the cluster. On the other hand,
if a large point mass, MBH , sits at the center, it will
cause the stars to form a density cusp, of characteris-
tic scale rcusp ∼ GMbh/σ2 (if σ is the random stellar
velocity).
• Results The results of this work is striking: essen-
tially every bulge yet observed (carefully enough) con-
tains a massive black hole. Furthermore, the mass of
the central object correlates tightly with σ, the veloc-
ity dispersion of the nearby stars. This is called the
Mbh − σ relation, usually quoted as
Mbh ∼ 1× 108(
σ
200 km/s
)x
M⊙.
The exact value of the exponent is still being argued
about: Kormendy (2001) gives x = 3.65, while Merritt
& Ferrarese (2001) give x = 4.80. I’d take x ∼ 4 as a
reasonable guess for now. About 40 galaxies had been
studied as of 2001 – about 2/3 of them by stellar cusps,
1/3 by gas disks – with BH masses extending from a
few million to a few billion solar masses. 3
1.6.2 Techniques: extend to AGN
The techniques just listed don’t work for most AGN,
due to the very bright nonstellar nucleus which
swamps the non-AGN signal. Two other techniques
are being developed. I personally don’t yet find them
as convincing as gas disks and stellar cusps, but they
have their adherents and the techniques are becoming
more reliable as time passes. They are:
• Reverberation mapping. Go back to our cartoon
of the emission line clouds close to the AGN. The line
widths tell us the velocity of the gas, which we assume
is gravitationally bound to the BH. To get its distance,
we look at variability. The gas emitting the broad lines
is photoionized by the central engine; when the ioniz-
ing flux varies, so will the ionization level in the line-
emitting gas, but after a delay due to the light travel
time. Monitoring of both the (ionizing) continuum and
3Comment from the author: this is very striking, and was totally
unexpected. Just about everyone in the field thought that only AGN
would have a massive BH in the core. A minor complication was
that quasars are much more common at high redshift – as discussed
below – so people were aware that there must be some ex-quasars
nearby. But few people suspected, before a few years ago, that
MDO’s would be so very common.
6 Physics 426 Notes Spring 2016
the emission lines can give us the distance of the line-
emitting gas from the BH. Combining this with the
linewidth gives us the mass of the BH. This has been
done for only a handful of AGN so far; the results (as
quoted in Merritt & Farrarese 2001) fit nicely on the
Mbh − σ curve determined for non-active bulges.
• X-ray line profiles. This attractive idea is still being
pursued observationally. The Kα line of iron has now
been seen in Xrays in several objects. The data suggest
it is very broad ( <∼ c/3 linewidth), and has “an asym-
metric red wing consistent with gravitational redshift”.
This is still a very new technique, and needs careful
modelling of the accretion flow in order to do anything
quantitative. Such work may be coming.
1.6.3 The galactic center: stellar orbits
The center of our galaxy is a special case, because it’s
so close (and easier to observe), and of course because
it’s of great personal interest to us. Different tech-
niques can be used here than in external galaxies, be-
cause we can see fainter objects and resolve smaller
scales. On the other hand, the GC is heavily obscured
as seen from here, so we can’t do anything optically.
Radio, IR and high frequencies (X- and γ rays) are our
tools.
As seen in the radio the region is quite a mess – a
complex distribution of thermal gas (HII regions), cold
molecular gas, and nonthermal emission (SNR and dif-
fuse). Most of this is extraneous to our focus here,
namely the existence and size of an MDO in the GC:
we need to search for a compact object and/or ordered
gas motions. Both things exist ...
• Streaming gas can be detected in radio continuum,
and its velocity measured with radio recombination
lines. Its structure has been called a “mini-spiral”, al-
though it is not as ordered as that name would suggest.
Its physical extent ∼ 1 − 22 pc, and its ordered rota-
tion speed ∼ 100− 200 km/s. If this gas is in ordered,
Keplerian motion it points to a gravitating mass of a
few ×106M⊙. This was the first strong sign, but the
uncertainty of the gas orbits and the lack of strong con-
straints on nuclear stars (could the mass be just a dense
star cluster?) kept the skeptical (like your author) from
accepting this as a detection of a BH.
• Sgr A* has long been known to be an unresolved ra-
dio source at what seems to be the dynamical center
of the galaxy. This argument is made as follows. The
data verify that Sgr A* is at the dynamical center of the
streaming gas ring/spiral, and also at the center of the
nuclear star cluster (described next). A more indirect
argument is it’s lack of random motion. VLB moni-
toring over 16 years showed that it has only the paral-
lax consistent with our motion around the GC; its own
space velocity can be no more than 15 km/s. This is
so much lower than other random motions that we can
infer it’s a massive object moving only very slowly. It
is unresolved, even at VLB scales4, making its physi-
cal size smaller than ∼ 1 AU. It is a compact, variable
synchrotron source: a small version of what we find in
other AGN. We don’t have the resolution at X- or γ-
rays to separate its high-frequency emission from the
general mess in the GC region; but an old report of an
e+ − e− annihiliation line, at 0.5 Mev, from the direc-
tion of the GC was tantalizing (and unfortunately has
never been repeated).
Thus: there is suggestive evidence of a massive, com-
pact thing in the GC. The recent result that tied this
down and convinced the skeptics is IR imaging of
the central star cluster. Individual stars can be distin-
guished easily within the central ∼ pc-sized star clus-
ter; and 10 years of monitoring (as reported by Schodel
etal, in a 23-author paper) allow measurement of the
stars’ proper motions. This is a great piece of work: the
orbits of individual stars were followed long enough
to track them through both peri-center and apo-center
passage, which allows a good determination of the or-
bital parameters and the mass of the central object.
This data fits well with other recent estimates of the
mass of the MDO, such as from velocity dispersions;
modelling seems to demonstrate robustly that the grav-
itating mass must be a point mass rather than a dense,
but extended, star cluster. The bottom line: the core of
our galaxy contains a MDO, almost certainly a BH, of
Mbh ∼ 3× 106M⊙.
References
Much of the general discussion is just “off the top of
my head”. Some useful general references on galaxies
are
• Binney & Merrifield, Galactic Astronomy
• Elmegreen, Galaxies & Galactic Structure
4That really means its angular size is small enough to be af-
fected by interstellar scattering, due to the radio waves propagating
through the turbulent ISM
Physics 426 Notes Spring 2016 7
• Sparke & Gallagher, Galaxies in the Universe
For more detail than you want on stellar dynamics and
the structure of galaxies, a very good book is
• Binney & Tremaine, Galactic Dynamics
I’ll put up AGN references later in the course. For now,
chapter 26 of Carroll & Ostlie is a nice introduction.
8 Physics 426 Notes Spring 2016
2 The Interstellar Medium
In the previous chapter we reviewed the gravitational
structure of bright (S and E) galaxies. But the stars
and dark matter are not all of the story. Each type
of galaxy contains gas as well as stars: this is the in-
terstellar medium (ISM). To set the stage, paraphrase
from Elmgreen, who’s talking specifically about our
galaxy1: “the ISM is like the ocean of a galaxy, a fluid
confined by gravity to a thin layer, and serving as a
reservoir for all of the material in stars and planets that
will ever form, evolve, and disperse.” The physics of
this ISM, and its connection to stellar birth and stel-
lar death, will be one of our main applications in this
course.
2.1 The diffuse ISM in our galaxy
What I call the diffuse ISM is the ISM that is truly “in-
terstellar” – sitting in the potential well of the overall
galactic disk, and not immediately involved with stars
or star formation regions.
2.1.1 How we oberve the ISM
Our understanding of the physical state of the ISM
has been driven by the data: recent work in radio as-
tronomy and high-energy astronomy has dramatically
changed the field. What are our current ways of look-
ing at the ISM?
• Optical: Gas with temperatures in the range roughly
103 to 104 K emits both continuum and spectral line
optical radiation, and cooler gas can be detected by the
absorption lines it produces when there is a continuum
optical source behind. We see interstellar absorption
lines, stellar reddening, diffuse Hα and other emis-
sion lines, and dark dust clouds. In addition “stellar
ISM” regions such as HII regions, supernova remnants
(SNR), and planetary nebulae emit optically. Galactic
dust tends to prevent us seeing through the plane of the
disk beyond a kpc or so, aside from special lines of
sight.
• Radio: the major discovery here was the HI 21 cm
line, which we see in absorption and emission. This
tracks atomic hydrogen, which is (of necessity) fairly
cool (T < 8000 K). In addition, the diffuse ISM is a
source of synchrotron radiation, which we see in the
radio. This latter comes from relativistic electrons (the
1in Burton, Elmegreen, & Genzel, eds., The Galactic Interstel-
lar Medium, SAAS-FEE Advanced Course 21, 1991.
cosmic ray population) interacting with the galactic
magnetic field.
• Infrared: another strong radiation source – one of
the strongest cooling mechanisms – is IR radiation
from dust grains. They absorb starlight and reradi-
ate it at temperatures ∼ 10 − 100 K. Dust exists in
many places throughout the ISM, at several tempera-
tures (“near IR”, “far IR”, etc.)
• Millimeter: most of the common molecules have
rotational transitions in the mm region (with the no-
table exception of H2); mm-wave astronomy is now a
powerful tool for studying star formation regions and
molecular clouds.
• Ultraviolet: this again typically samples only the lo-
cal region, <∼1/2 kpc, due to obscuration. At this band
we see hot gas, 105 − 106 K, from the so-called “local
bubble”; in absorption, there are important transitions
of atomic and molecular hydrogen. With the advent of
UIT, the UV is also becoming a nice tool for studying
the ISM in external galaxies.
• X- and γ-rays: the local bubble also emits soft X-
rays. In addition, the galactic plane is a source of
harder X-rays and γ-rays; some of these come from
hot gas (such as in SNR), and the hardest component
comes from nuclear reactions between the cosmic rays
and the ISM.
• Plasma propagation: radio signals are affected by
propagation through an ionized plasma. In addition,
the ionized plasma can emit thermal bremsstrahlung.
We can use the following three means to measure elec-
tron densities, path lengths and magnetic field:
dispersion measure : DM ∝∫
nedl
emission measure : EM ∝∫
n2edl
rotation measure : RM ∝∫
neB · dl
(2.1)
The first refers to plasma dispersion, the fact that
the phase speed of an EM wave depends on its fre-
quency. The second measures the emissivity due to
bremsstrahlung radiation. The third refers to Faraday
rotation, the rotation of the plane of polarization due to
passage through an ionized, magnetized plasma.
2.1.2 A multi-phase equilibrium?
From such observations, we find that the diffuse ISM
is very complex. It comes in several phases, with com-
Physics 426 Notes Spring 2016 9
plex spatial structure. Different authors classify the
phases differently, and papers on this topic can contain
a flurry of acronyms (CNM, WNM, WIM, MOWIM,
RWIM, HIM, etc, etc, etc). I summarize as follows.
• Neutral gas refers to atomic hydrogen, HI. (Molecu-
lar H2 and other species are found in self-gravitating
clouds, thus are discussed in the next section; they
are not really part of the diffuse ISM). We now know
that the spatial structure of neutral HI is quite compli-
cated. First, there is “cold” HI (seen in absorption) and
“warm” HI, still neutral, seen in emission. Heiles esti-
mates the warm HI has T >∼ 1000 K, n ∼ 0.3 cm−3
The cold phase comes in sheets, filaments, and bub-
bles. Early observations – which had only spectral
resolution (and so picked up gas at different veloc-
ities), not spatial, talked about “interstellar clouds”.
This terminology is still around, but our cartoon is
no longer a raisins-in-a-pudding picture. Typical tem-
peratures are ∼ 10 − 75 K and typical densities are
∼ 20 − 2500 cm−3. There seems to be a wide range
of “cloud” sizes (i.e., path lengths through clumps or
sheets), up to big “clouds” >∼ 100 M⊙.
• Warm ionized gas refers to partly to mostly ionized
hydrogen. This used to be the “intercloud medium”;
then for awhile it was thought to be located on in-
terfaces between the cold HI clouds and the really
hot, coronal gas. Now, observations of external galax-
ies show that there is also a diffuse, extended warm
component, occupying filaments, clouds, bubbles and
chimneys. We see it mainly in diffuse Hα and other
optical lines; earlier work also found it locally in ab-
sorption lines.
Heiles argues that these are two separate types of
warm, ionized gas. One is the diffuse component,
throughout the galactic plane, maintained by photoion-
ization by starlight. Heiles estimates T ∼ 8000 K
and n <∼ 0.1 cm−3 for this gas (sometimes called the
Reynolds layer, after the person who first studied it in
depth). The second type is, indeed, the warm interfaces
between cold and hot gas. Heiles gives T ∼ 8000 K
and n ∼ 0.1− 0.4 cm−3.
• Hot “coronal” gas refers to the phase at T ∼ 105 −106K, n ∼ 10−3 − 10−2cm−3. We detect this gas by
its X-ray emission, and also some UV lines. It is as-
sociated with, but not confined to, the interiors of su-
pernova remnants and “superbubbles” (large, multi-SN
complexes).
If we look at all phases, we see that each has the prod-
uct nT on the order of a few thousand (cm−3K). Thus,
within the accuracy of the diverse data, each has the
same typical pressure: this product converts to an en-
ergy density ∼ 1 eV/cm−3. We are seeing three phases
in approximate pressure balance with each other.
2.1.3 Other components of the ISM
There are three other significant components of the dif-
fuse ISM.
• Dust grains tie up about 1% of the mass of the
ISM, including most of the heavy elements. They are
often found with, or comprised of, complex organic
molecules. Polycyclic aromatic hydrocarbons (PAH’s,
the stuff of soot) are thought to be one major group.
Dust absorbs and scatters optical starlight very effec-
tively (this is why we can’t see vary far optically) and
reradiates it in the infrared, providing a major part of
the bolometric luminosity of the galaxy.
• Cosmic rays are relativistic particles, electrons and
ions, tied to the galaxy by its magnetic field. We de-
tect them directly at the earth though their distribution
is modified by passage through the solar wind and the
earth’s atmosphere. Their detected energies range from
∼ 1010 eV to ∼ 1021 eV. We also detect them in-
directly by their diffuse synchrotron emission (in the
galactic magnetic field) and by their high-energy pho-
ton emission (arising from nuclear reactions with the
thermal ISM). Their energy density locally is compa-
rable to that of the ISM gas, ∼ 1 eV/cm−3, and may
increase by a factor of a few going inwards towards
the center of the galaxy.
• the Magnetic field of the galaxy is detected through
Faraday rotation, synchrotron emission, optical po-
larization of starlight (due to alignment of IS grains
across B) and Zeeman splitting. It is fairly ordered,
with field lines tending to lie in the galactic plane, and
somewhat along the spiral arms. Typical field strength
is usually quoted as ∼ 3µG, with strong spatial vari-
ation, and higher fields locally in some regions. Its
energy density locally is also – you guessed it – >∼1 eV/cm−3. As with cosmic rays, there is some sug-
gestion that the energy density in the field rises going
towards the galactic center.
2.2 “Star stuff” in our galaxy
In addition to the diffuse medium, many well-known
examples of so-called interstellar matter come from
ISM closely related to stars. In particular, any of the
10 Physics 426 Notes Spring 2016
prettiest pictures come from nebulae associated with
young stars or old, dying stars. They include:
• HII regions are regions of ionized hydrogen sur-
rounding, and ionized by, hot young stars. They are
strong optical and radio sources.
• Planetary nebulae are the outer layers of older stars,
ejected by instabilities in the star’s structure.
• Stellar winds are produced by nearly all stars at
some level; hot, young stars have by far the strongest
winds. While not as spectacular as the first two, winds
are strong sources of mass and energy supply for the
ISM.
• Stellar jets are produced both by young stars, in
the formation process, and by compact stellar rem-
nants (e.g., neutron stars and their surrounding accre-
tion disks). They are striking radio and optical sources
when we can catch them; a few have apparent superlu-
minal expansion velocities.
• Supernovae remnants result when massive stars
eject something like half their mass, or more, in a vi-
olent explosion. We see SNR as strong radio, optical
and X-ray sources; in addition they are thought to be
strong sources of cosmic rays. They are also important
contributors to the energy and mass budget of the ISM.
These nebulae all share the property that they are over-
pressured relative to the diffuse ISM: they arise from
and are tied to stars. In addition, molecular gas is found
in overpressured, self gravitating clumps called
• Molecular Clouds. These are detected in “trace”
heavy-element molecules such as CO, which have
strong millimeter line transitions; we infer that H2 is
also present, in larger quantites. These clouds can
be very cold, T ∼ 30 − 100K, and very big, up to
∼ 105 − 106M⊙. They have line widths much greater
than the Doppler width one would expect from their
temperatures: thus they are either collapsing (under
their own self-gravity) or supported by turbulent, dis-
ordered internal motions. They are the stellar nurseries
of the galaxy, the sites of ongoing star formation.
2.3 Galactic ecology
Taking the galaxy as a whole, we must keep in mind
that the ISM is not static. It is continually being con-
sumed in new star formation (at a rate ∼ 3− 10M⊙/yr
over the galaxy) and continually being replenished by
stellar ejecta (everything from winds to supernovae).
It loses energy by radiation, over all wavelengths, and
gains energy as it is replenished by the stellar ejecta.
Thus, we can think of stars as simply long-lived phases
of the ISM; they are formed from it and they return to
it.
However, there is a trend: not all stellar material is
returned to the ISM. Low-mass stars become white
dwarfs, and quietly cool forever; high-mass stars recy-
cle much of their mass, but most are believed to leave
compact cores, neutron stars or black holes. Thus, the
overall trend of the system is towards cold, dense ex-
stars. This will take awhile, however; at present we
find quite a mixture of stellar and diffuse matter in the
galaxy.
2.4 The ISM in Ellipticals
Our understanding of ellipticals (the stars as well as the
ISM) has changed dramatically over the past couple of
decades. Originally it was thought that these galaxies
have no ISM. The older stellar population characteris-
tic of E’s, and the lack of strong star formation, also
seemed to argue against any interstellar matter. Then
people started looking...and it now appears that these
galaxies also have a complex, multi-phase ISM, with
total mass comparable to that in spirals. Unlike spirals,
however, the ISM in E’s tends to be hotter, that is with
smaller amounts in cool, neutral or “warm” phases, and
most of the gas being in the hot phase. This is still a
new and evolving field, being limited both by detector
sensitivity and by amount of effort (not as many peo-
ple look at these faint, distant things). In these notes I
summarize the current state of things.
2.4.1 Everything that isn’t the hot phase
Some – not all – E’s have now been shown to have
neutral hydrogen, molecular gas, dust grains, and a
“warm” ISM (the latter detected in optical emission
lines, placing it at ∼ 104K). Some have been shown
to have onging star formation in their cores. Because
a smaller fraction of the total ISM is “cool”, its sig-
nal is faint; a non-detection does not necessarily mean
there is no cool gas in a particular galaxy. On the
other hand, there seems to be an ongoing argument
as to whether galaxies detected strongly in, say, HI or
CO are “typical” (“real ellipticals don’t eat quiche,” to
quote Rupen2). Some authors argue that only unusual,
or disturbed, E’s contain detectable amounts of cool
gas...thereby defining a “pure E” as one without such
2M. Rupen, private communication, a few years back.
Physics 426 Notes Spring 2016 11
gas. Given sensitivity limitations, I tend to think this
view may eventually be proved wrong. But I agree that
we do not, yet, understand the multi-phase cool gas in
the ISM of an elliptical.
2.4.2 The hot phase – the x-ray loud gas
Our understanding of the ISM in E’s started to change
in 1985, when the ⁀Einstein X-ray observatory discov-
ered that normal E’s are strong X-ray sources. (This
has since been explored in detail by ROSAT, and now
CHANDRA is adding to the picture). “Hot” means
at temperatures to emit X-ray bremsstrahlung: T ∼107K. (We can estimate the temperature roughly from
the fact that the gas is an X-ray source, and more
specifically from X-ray emission lines). The spatial
distribution of the gas is consistent with it sitting in hy-
drostatic equilibrium in the potential well of the galaxy.
The amounts are large: Sarazin gives 109 − 1011M⊙
in gas, or comparing to the optical (blue) light of the
galaxy, Mgas/M⊙ ≃ 0.2(LB/Lsun). This is most of
the ISM; the fractions estimated in all of the cool com-
ponents are much smaller.
Why is so much of the ISM hot in an elliptical? Think
about the ecology of this ISM. As in a spiral, the
gas is ejected from stars, by the usual mass-loss pro-
cesses (stellar winds and SNe). Unlike a spiral, how-
ever, the stars have quite high random motions (200-
300 km/s is typical for gravitational support). It fol-
lows that collisions between the ejected gas, and ei-
ther gas ejected by nearby stars or the local ambient
ISM will thermalize the kinetic energy of the injected
gas. (This is a direct application of shock physics —
which we’ll see later in the course). This means the
gas is effectively injected with at least T ∼ mσ2/k ∼7 × 107K(σ/300 km s−1)2 (any larger injection ve-
locities, due to winds or SN flows, will only raise this
number). We can also note that the radiative cooling
from this hot gas will be less effective than from the
cooler, denser gas of a spiral galaxy. The details of
why this is so must wait until later in the course; it
has to do with lower density (the same amount of gas
spread over a larger volume) and strongly ionized gas
at these hot temperatures (so emission lines, which are
strong coolants, aren’t as important).
References
Much of the general discussion is just “off the top of
my head”. Useful recent references come from meet-
ing proceedings:
• Arnaboldi, Da Costa, & Saha, eds, 1996, The Second
Stromlo Symposium: the Nature of Elliptical Galaxies.
Several articles; especially that by Sarazin on the X-ray
loud gas in E’s.
• Duric, 1999, in New Perspectives on the ISM, eds.
Taylor, Landecker & Jones; p. 161 – on cosmic rays
and galactic B field.
• Sadler, 2002, in Da Costa & Jerjen, eds., The Dy-
namics, Structure & History of Galaxies, p. 215.
• Woodward, Bicay & Shull, eds, 2001, Tetons 4:
Galactic Structure, Stars, & the ISM; especially the ar-
ticles by Heiles & Dickey.
12 Physics 426 Notes Spring 2016
3 Some Radiation Basics
In this chapter I’ll store some basic tools we need for
working with radiation astrophysically. This mate-
rial comes directly from Rybicki & Lightman (“RL”),
where you can find a more complete discussion of it
all.
Our apporoach will be ray optics. You remember that
radiation can be approached as EM waves, as discrete
photons, or in the ’ray optics’ limit – we’re thinking
in terms of ray optics here. Some of this material will
look pretty dry to you ... as you go along, look for these
important quantities, which will be useful tools for us
later.
Intensity, Iν (equation 3.1)
Flux, Fν (equation 3.2)
Energy density, uν (equation 3.4)
Intensity in thermal equilibrium (TE), Bν
(equation 3.17)
Emission coefficient (per solid angle), jν(equation 3.23)
Emissivity, ǫν → 4πjν ( equation 3.24)
Absorption coefficient, κν (per solid angle;
equation 3.26)
Source function, Sν = jν/κν (equation
3.29)
Opacity, τν ( equation 3.27)
OK, here goes.
3.1 Radiation: some important definitions
We begin with some important definitions.
• Intensity. Consider a little (differential) area dA =dA n, with a radiation beam passing through it. At a
particular frequency ν, the energy passing through dAin a particular direction (θ, φ) (measured relative to n),
per frequency range dν, per time dt, and per solid angle
dΩ, is given by
dE = Iν dA dt dν dΩ. (3.1)
This relation implicitly defines our basic quantity, the
intensity: Iν1 Most of our other quantities are defined
1NOTATION ALERT: Iν is traditional notation in this field,
and means “I is a function of ν”. So, I(ν) ↔ Iν , and ditto for
jν , κν , etc.
in terms of Iν . Intensity is also called specific intensity,
brightness or surface brightness. In cgs, its units look
like erg cm−2 Hz−1 s−1 str−1.
• Flux is the net energy passing through dA, in all di-
rections:
Fν =
∫
Iν cos θ d(cos θ) dφ (3.2)
(with units erg cm−2 s−1 Hz−1 or W m−2 Hz−1).
• Heads up here: Flux and intensity are
similar but not the same; you’ll need to be
able to work with both. There is some de-
tailed discussion in Appendix I to this chap-
ter. Basically, intensity is what is shown in
an image of a source and flux is what you
get if you integrate everything in the image.
• One can also define a mean intensity, averaged over
all solid angles:
Jν =1
4π
∫
Iν d(cos θ) dφ (3.3)
• The energy density is clearly related to the intensity
by a factor of lightspeed. The most useful definition is
in terms of the angular mean.
uν =4π
cJν =
1
c
∫
Iν d(cos θ) dφ (3.4)
The units are erg cm−3 Hz−1 (of course).
• Frequency integrated. You should also note that
each of the quantities above can be integrated over fre-
quency:
F =
∫
Fνdν ; I =
∫
Iνdν ; u =
∫
uνdν
(3.5)
and so on.
3.2 Thermal equilibrium: an ideal gas
You’ve probably seen this elsewhere; I’ll just review
the basics here. Consider a small system – say, one
atom in a gas – which is in thermal contact with a large
system. That means that energy exchange is allowed,
most likely through collisions with the rest of the gas.
Let the big system – the so-called “reservoir”– have
a temperature T . The fundamental result of classical
Physics 426 Notes Spring 2016 13
thermodynamics is that the probability of finding the
small (test) system in a state of energy E is
P(E) ∝ e−E/kBT (3.6)
This is the Boltzmann factor; the proportionality con-
stant is used to normalize the probability, in a specific
system (as, below).
Now, let’s apply this to an ideal, monatomic gas. Let
the test system be a single atom in the gas, and let the
rest of the gas be the reservoir, at T . Each atom has
a mass, m, and a random velocity, v; the energy as-
sociated with this velocity is E(v) = 12mv2. (This
could of course describe a plasma as well as a neutral,
atomic gas). The probability that a particle has energy
E is then
P(v) ∝ e−mv2/2kBT (3.7)
We want to use this to derive the distribution of parti-
cle velocities. In addition to the Boltzmann factor, we
need to know the number of ways in which a particle
of velocity v can have E(v). If the gas is isotropic
– if there are no restrictions on the possible orien-
tation of the velocity vector – then the factor which
weights the Boltzmann factor is just the number of pos-
sible directions the velocity vector can point. That is,
P(v) ∝ 4πv2e−mv2/2kBT . Now,if we normalize the
answer to the total number density, so that
n =
∫
v
P(v)d3v =
∫ ∞
0f(v)dv (3.8)
defines the distribution function, f(v), we end up with
f(v) = 4πn
(
m
2πkBT
)3/2
v2e−mv2/2kBT (3.9)
which is (one way to write) the Maxwell-Boltzmann
distribution of particle velocities.
Taking means (or moments) of this distribution, we
find the mean particle velocity,
〈v〉 =∫ ∞
0vf(v)dv =
(
8kBT
πm
)1/2
(3.10)
and
〈E〉 =∫ ∞
0
1
2mv2f(v)dv =
3
2kBT (3.11)
We can extend this type of analysis to find the pressure
of the gas. For a simple, MB distribution, this recovers
the usual ideal gas law:
p = nkBT (3.12)
This analysis can also be used to get the pressure of a
relativistic ideal gas, or of a photon gas. I’ve put the
details in Appendix II to this chapter.
3.3 Thermal equilibrium: radiation
Here, we consider a photon gas which is in thermal
contact with something at temperature T . That “some-
thing” could be the walls of a closed box (the prover-
bial black body), or a dense gas cloud (which could be
a dark interstellar cloud, or a star). As with a parti-
cle gas, thermal contact means the photons exchange
energy with the “something” through collisions; here,
this could mean particle-photon scattering, such as
Thompson scattering of photons on electrons; or it
could mean that the matter absorbs and re-emits pho-
tons. Now, in particle-particle collisions the particle
number is conserved (in standard, elastic collisions,
anyway). In photon-matter “collisions”, on the other
hand, the photon number is not conserved; it is quite
possible for an atom to absorb one photon and re-emit
several, still while conserving energy (for instance, the
atom could absorb to the n = 10 level and re-emit
10 → 8, 8 → 5 and 5 → 1 photons as it decayed).
Thus, in dealing with the particle gas, we normalized
the probability function by assuming a certain total
number density of particles. For radiation in TE, the
number density of photons is predicted if we know T .
The analog of the Boltzmann factor for a photon gas
is the Planck distribution: the probability of finding a
photon at energy E = hν is
P(E) ∝ 1
eE/kBT − 1(3.13)
which, of course, resembles the classical (non-
quantum mechanical) Boltzmann factor for energies
E >> kBT . We find the density of states allowed
at energy E just as we did for particles, by looking
at the number of ways photons of wavevector k =p/h = E/hc = 2πν/c = 2π/λ can be put into a vol-
ume. But, recalling standing waves, we remember that
a one-dimensional box of length ℓ can contain standing
waves of wavenumber k = 2πq/ℓ if q is any integer.
So, the number of photon states in three dimensions in
(k,k+ dk) is
d3q =ℓ3
8π34πk2dk (3.14)
From this, we find the density of states per volume by
dividing by ℓ3 and adding the usual factor 2 for the two
14 Physics 426 Notes Spring 2016
spin (polarization) states, and express things in terms
of ν rather than k:
density of states = 8πν2
c3dν (3.15)
This is the factor which weights P(E) (from equation
3.6) to find the density of photons at energy E = hν.
(Note that this is 4π times larger than RL’s expression
for ρs; they are working in density of states per solid
angle.) However, it is common to multiply this density
by hν to find the energy density of radiation in TE at
T :
urad(ν, T )dν =8πν2
c3hν
ehν/kBT − 1dν (3.16)
(In dealing with black body radiation, watch out for
units; different authors do different things. urad in
equation (3.16) has units energy/volume-Hz; some au-
thors divide by 4π to get energy/steradian-volume-Hz.)
Another way to express this result is in terms of the en-
ergy in radiation crossing a unit area per Hz per time
(and usually per steradian). As we saw above, this
quantity is the intensity, and is related to the energy
density in radiation by I(ν) = curad(ν)/4π. For (and
only for) the specific case of Black Body radiation, the
intensity is denoted B(ν, T ) or Bν(T ), rather than Iν .
For black body radiation, we therefore have
B(ν, T ) = Bν(T ) =2hν3
c21
ehν/kBT − 1(3.17)
with units, energy/area-time-steradian-Hz. If you are
working with wavelengths rather than frequencies, the
analogous version is
B(λ, T ) = Bλ(T ) =2hc2
λ5
1
ehc/λkBT − 1
and it will give you an intensity in units of energy/area-
time-steradian-wavelength.
Equations (3.16) and (3.17) are the basic result describ-
ing radiation in TE. However, several extensions and
approximations are standard. First, we can integrate
them over frequency to find the total energy (per vol-
ume, or per area per time):
urad(T ) =
∫ ∞
0urad(ν, T )dν
=4π
c
∫
Bν(T )dν = aT 4(3.18)
where the constant a = 8π5k4B/15c3h3 = 7.56 ×
10−15 erg/cm3deg4. In addition, we have
B(T ) =
∫ ∞
0Bν(T )dν =
ac
4πT 4 (3.19)
and, finally, the emergent flux obeys F = πB(T ), so
that
F (T ) =
∫
Fνdν = π
∫
Bνdν = σSBT4 (3.20)
where σSB = ac/4 = 5.67 × 10−5erg/cm2deg4s−1.
Finally, a couple of limits of Bν(T ) are worth noting.
For high frequencies with hν ≫ kBT , we have the
Wien limit:
Bν(T ) ≃2hν3
c2e−hν/kBT (3.21)
which recovers the exponential form. For low frequen-
cies hν ≪ kBT , we have the Rayleigh-Jeans limit:
Bν(T ) ≃2ν2
c2kBT =
2
λ2kBT (3.22)
which is very frequently used as an approximation to
the black body function in the radio part of the spec-
trum.
3.4 Radiative transfer
Start with a beam of radiation, described as usual by
intensity Iν . Consider such a beam, from some back-
ground source, hitting a slab of material. It will be
absorbed by the material as it passes, and the material
may well also emit radiation into the beam.
x=0 x=L
I o
I
Figure 3.1 Geometry of radiation transfer through a slab
of material, possibly with a background source Io.
3.4.1 More definitions
We need more definitions now. In addition to the in-
tensity, Iν , we need to think about the plasma through
which the beam passes.
Physics 426 Notes Spring 2016 15
• The plasma has an emission coefficient jν , defined
in terms of the contribution to a radiation beam (Iν as
it propagates a distance dx:
dIν = jνdx (3.23)
and the units of jν are erg cm−3 s−1 Hz−1 str−1.
This is the fundamental radiated power, at frequency ν,
from the matter; its details depend on the local physics.
Note also that the mean intenstiy Jν that we saw earlier
is not the same as the emission coefficient jν .
• For some problems it’s more useful to work with the
volume emissivity,
ǫν = 4πjν (3.24)
(This assumes isotropic emission). We’ll run into this
later.
• We also need the absorption coefficient κν ,2 which
describes the fractional absorption or scattering of a
radiation beam, per unit length dx:
dIν = −κνIνdx (3.25)
This has units cm−1. It can also be written microscop-
ically (to reveal the physics), in terms of the number
density of absorbers n and their absorption cross sec-
tion σν : κν = nσν . Question for the reader: How,
then, is the absorption coefficient related to the mean
free path of a photon?
3.4.2 Transfer analysis
With these definitions, the basic transfer equation can
be written down,
dIνdx
= jν − κνIν (3.26)
Before solving this, we introduce an important and
useful quantity, the optical depth:
τν =
∫ ℓ
oκνdx (3.27)
where the integral is taken from back to front through
the slab of matter. From the discussion of κν , above,
we see that τν = ℓ/λ measures the number of absorp-
tion mean free paths through the source. We would
expect, then, that a system with τν ≪ 1 would have
2NOTATION ALERT: about half the literature uses αν for the
absorption coefficient; the other half uses κν , as I do here.
little effect on any source behind it (that is, it would be
nearly transparent), and a system with τν ≫ 1 would
be nearly opaque, absorbing most of the light. We can
rewrite (3.26) with τ as the independent variable:
dIνdτν
= Sν − Iν (3.28)
where we have defined the source function,
Sν =jνκν
(3.29)
Now, solve (3.28). If we put a source of intensity Iobehind the slab, the formal solution (remember inte-
grating factors?) is
Iν(τν) = Ioe−τν +
∫ τν
0Sν(τ
′)e−(τν−τ ′)dτ ′ (3.30)
Look at this: the first term is simply the attenuation of
the background source by the slab. The variable τ ′ is a
distance through the slab, but it’s measured in dimen-
sionless optical depth units. The second term describes
the emission of radiation from within the slab, at po-
sition τ ′, and the attenuation of this radiation by the
smaller optical depth, τν − τ ′, between the emission
point and the front of the slab.
3.4.3 Optically thick and thin limits
In the important case of a homogeneous source, with
Io = 0, (3.30) simplifies to
Iν(τν) = Sν
(
1− e−τν)
(3.31)
describing emission only from the cloud/slab itself.
This has two important limits:
• Optically thin, τ ≪ 1: we see
Iν ≃ Sντν = jνℓ (3.32)
This limit just integrates the emissivity through the
cloud, without modifying it by internal absorption.
• Optically thick, τ ≫ 1:
Iν ≃ Sν (3.33)
In this limit, the emergent intensity is just equal to
the source function. NOTE this may be quite differ-
ent from the internal emissivity; transfer through the
source has modified the spectrum.
From the optically thick limit we can make a couple of
important connections.
16 Physics 426 Notes Spring 2016
• Consider the case when radiation is in TE with the
local plasma. This means Iν → Bν (the Planck func-
tion, (3.17 or its limits); and also τν ≫ 1 (because
you need lots of collisions, it i.e. optically thick, to
gain TE). Thus, from (3.31), we expect Sν ≃ Bν (the
source function approaches the Planck function), and
from this we derive an important relation:
jν ≃ Bνκν (3.34)
This is called Kirchoff’s law. It says: if we can assume
thermal equilibrium (as we will do in chapters 3 and
4), the emissivity and absorption are related through
the Planck function.
• Astronomers working at radio frequencies com-
monly quote the intensity, Iν , in terms of the Bright-
ness Temperature, TB , defined by
Iν = 2ν2
c2kBTB (3.35)
But from comparing (3.22) and (3.35), we find that
TB → T as the source becomes optically thick: the
brightness temperature approaches the physical tem-
perature. (Question for you: if the source is optically
thin, how are T and TB related? How does your an-
swer depend on the conditions in the source?)
3.5 Appendix I: some examples with intensity
Working with intensity, flux, etc, can be confusing (at
least to your author!); so I’m putting some specific ex-
amples here – mostly directly from RL.
3.5.1 Isotropic radiation field
If the radiation field is isotropic life is simple: Jν = Iν(is that obvious?) and Fν = 0 (there’s no net energy
flow; there’s as much going “out” as going “in”). Also,
by inspection of (3.4), we see that uν = 4πIν/c.
3.5.2 Intensity is constant along a ray
We should note one important fact: in the absence
of absorption or emission, the intensity Iν is constant
along any ray. RL present one (rather formal) deriva-
tion of this, illustrated in Fig 3.2, which I summarize
here. The key points are that intensity is defined per
solid angle, and that energy is conserved. Think about
the set of rays passing through both dA1 and dA2. The
energy in that set of rays is
dE = I1 dA1 dt dΩ1 dν = I2 dA2 dt dΩ2 dν (3.36)
But, thanks to the inverse square law, dΩ1 = dA2/R2,
and dΩ2 = dA1/R2. Thus, because the same dE
passes through both little areas, we must have I1 = I2.
Q.E.D.
2
R
dA1 dA
2
dΩ 1 dΩ
Figure 3.2 One way to establish the constancy of I along a
ray. Two little areas, dA1 and dA2, are separated by R. dA1
subtends a solid angle dΩ2, as seen by 2; and vice versa for
dA2 as seen by 1. Following RL Fig 1.5.
Wait .. does this seem unphysical? What about the
inverse-square law that we know applies to radiated
power? The key is that intensity is not the same as flux
– and flux satisfies the 1/R2 law. This is contained in
(3.36), because the solid angle dΩ = dA/R2; so that
the energy per area passing through any dA falls off
∝ 1/R2.
Or, if this argument isn’t very transparent, you might
prefer the next example.
3.5.3 Flux from a sphere
Here’s a nice example, to verify that radiative flux does
indeed obey the inverse square law. Put yourself at dis-
tance D from a sphere of uniform brightness B; that
means all rays leaving the surface have the same inten-
sity, I = Io, independent of direction. (The geometry
is shown in Figure 3.3).
(I = B)
R
D(obs)
θ
θc
Figure 3.3 The geometry used for calculating the flux
from a uniformly bright sphere. The observer is at a dis-
tance D away; the sphere radius is R; the radius subtends
an angle θc as seen by the observer; the intensity is assumed
uniform, I = Io, along every ray that leaves the surface.
Following RL Fig 1.6.
The intensity you see, then, is I = Io from rays which
intersect the sphere; and I = 0 from other angles. The
Physics 426 Notes Spring 2016 17
flux you observe is thus
F =
∫ 2π
0dφ
∫ θc
0I(θ, φ) cos θd cos θ (3.37)
But this is easy to integrate:3
F = πIo(
1− cos2 θc)
= πIo sin2 θc (3.38)
This nicely recovers the inverse square law as long as
the distances involved are not cosmological:
F = πIoR2
D2(3.39)
for a uniform source of circular cross-section. Also,
note that at R = D (when you’re right at the surface of
the star), the flux is
F (D = R) = πIo (3.40)
We can also invert this solution. Say we observe flux
F at earth.4 The intensity at the source is, clearly,
Iν =Fν
π
D2
R2(3.41)
But also, remember θc = sin−1(R/D) ≃ R/D (this
last for a distance source); so the solid angle subtended
by a distant source is Ωc = πθ2c . Thus, we can go from
the flux (at earth) to the intensity (at the source) by
Iν =Fν
Ωc(3.42)
3.6 Appendix II: more on pressure
3.6.1 Ideal gases: the pressure integral
We can also use the MB distribution to find the pres-
sure of an ideal gas. (We will derive this is a fairly
formal way, to use later on). (For this subsection, we
use p for the single-particle momentum, and P for the
pressure). The pressure is, of course, the force exerted
per unit area from collisions of the gas particles. Con-
sider some “test surface” within the gas, and we can
find this force. One particle, with momentum p, ap-
proaches this surface at angle θ. When it recoils from
3Just to be more difficult ... some authors (e.g. Mihalas “absorb
the π (in 3.39) into the definition of flux”, and call it “astrophysical
flux”. I think the moral is, be careful when you go from author to
author – JAE.4NOTATION ALERT: the flux at earth is often called S or Sν ,
at least in radio astronomy.
the surface, it transfers momentum ∆p = 2p cos θ to
the surface. Now, the rate of particles approaching at
this p and this θ is
j(p, θ) = v(p) cos θ1
2f(p) sin θ
(that is, the normal velocity times the number of parti-
cles “at θ”; and noting that only half of the particles “at
θ” are approaching the surface). The pressure, then, is
this rate times the ∆p per collision, integrated over all
angles and all velocities:
P =
∫ π/2
0dθ
∫ ∞
0dp2p cos θj(p, θ) (3.43)
If we now assume the gas is isotropic, we can do the θintegral right away, and we end up with
P =1
3
∫ ∞
0pv(p)f(p)dp (3.44)
If we put in the MB distribution, from (3.9), and as-
sume a subrelativistic gas, so that v(p) = p/2m, we
find P ∝∫∞0 p2e−p2/2mkBTdp, and end up with
P = nkBT (3.45)
as we expect.
Another interesting limit is that of a relativistic ideal
gas, in which the single particle energy E ≫ mc2. In
this limit, we have p ≃ E/c and v ≃ c, so that
P ≃ 1
3
∫
Ef(E)dE (3.46)
(where we have used f(E)dE = f(p)dp). But the
integral is just the energy density of the gas, u; so we
have
P =1
3u (3.47)
which is a general result for an internally relativistic
gas (whether of particles or of photons).
3.6.2 Radiation pressure
Although not limited to radiation in TE, this is as good
a place as any to put the basic facts about radiation
pressure. For a general radiation field, we can use the
general equation (3.43), with the photon flux written as
j(ν, θ) =c
2hνcos θ urad(ν, θ) sin θ (3.48)
18 Physics 426 Notes Spring 2016
To evaulate Prad, we would have to know the angular
distribution of urad (for instance, the basic radiation-
pressure applications, such as an astronaut with a flash-
light, or a spaceship with a light sail near the sun, as-
sume a highly directed urad).
One simple limit is the case of an isotropic radiation
field (which is a good description of a radiation field
in which the photons scatter, such as the interior of a
star – that is, a photon field which is probably also in
TE). Here, we can start with equation (3.44), and we
can certainly use the relativistic limit; thus, we get
Prad =1
3urad (3.49)
in general, for an isotropic radiation field.
The other simple limit is the case of a unidirectional
radiation field – for instance radiation from the sun
as seen at the distance of the earth. We could more
properly talk about radiation force here: dimension-
ally that’s the (radiation power)×(surface area of the
absorber)/c. One application of this is the luminos-
ity at which the radiation pressure (or force) from
some object of mass M balances its gravity. This is
the Eddington luminosity. If we’re talking about ion-
ized gas for which Thomson scattering (cross section
σT ) dominates, the Eddington luminosity is Ledd =4πcGMmp/σT (which you remember from last term).
References
The discussion about ideal gas laws, pressure integrals,
etc, can be found in any basic statistical mechanics
book – two good ones are
• Reif, Statistical and Thermal Physics; Kittel, Ther-
mal Physics.
The material on radiation comes straight from one of
the fundamental references in the field,
• Rybicki & Lightman, Radiation Processes in Astro-
physics
but also
• Mihalas, Stellar Atmospheres, has a good discussion
of the intensity basics.
Key points
• Basic definitions, Iν ; Fν ; uν , and what they mean;
• Radiation in TE: Black body physics
• Radiative transfer: jν , κν , and τν : what they are,
what they mean.
• Radiative transfer: solutions to Iν(τν), optically
thick and thin limits.
• Brightness temperature and the approach to TE (as
τν gets big).
Physics 426 Notes Spring 2016 19
4 Bremsstrahlung radiation
Bremsstrahlung arises when a free electron is accel-
erated in the field of an ion – hence the English name
“free-free,” representing the transition between two un-
bound electronic states. The German name “braking
radiation” refers to the acceleration of the electron.
4.1 Some basic tools
Before we start bremsstrahlung per se, we need to in-
troduce two important general tools used for general
analysis of astrophysical radiation.
4.1.1 Power; Larmor formula
You probably remember that if you shake an electron,
it will radiate E&M waves. We want to connect the to-
tal power in the E&M radiation to “how hard the elec-
tron was shaken”.
The formal result is that a charge e, which feels an ac-
celeration a(t), radiates a power (erg s−1) given by
cgs : P (t) =2
3
e2
c3|a(t)|2 (4.1)
Note this is cgs.1
To derive this, you need to work out the E and B
fields which are produced by the accelerated charge;
then fold them together into the Poynting flux, S =cE×B/4π.2 That gives us the energy per unit area per
unit time carried away from the particle by the fields,
i.e. the radiated power. Griffiths has a good derivation
in chapter 11; Rybicki & Lightman have a more terse
derivation in chapter 3. You should note that this for-
mula holds for non-relativistic motion; we’ll extend it
to the relativistic case, later.
4.1.2 Spectrum: Fourier analysis
The other basic tool we need is the spectrum of the
radiation. When our electron is shaken, it emits a pulse
of radiation, which has a finite duration. This pulse
is a wave packet – a superposition of E&M waves of
various frequencies. The pulse width in time, ∆t, is
related to the range of frequencies in the packet, δν, by
1In SI, the formula is
SI : P (t) =µo
6π
e2
c|a(t)|2
2or, S = E×B/µo in SI.
the usual uncertainty principle: ∆t∆ν ∼ O(1) (as in
Figure 4.1). The amplitude of frequency component νgives what we’ll identify as the radiation spectrum.
E(t)
∆t
t
Figure 4.1 Remembering the wave content of a wave
packet. The time duration of this packet ∼ ∆t (note this
is estimated “by eye” here); it contains frequencies <∼1/∆t.
To get there formally, we use the Fourier transform
(“FT”). I take this from chapter 2 of Rybicki & Light-
man. Think about our pulse of radiation; let the electric
field in the pulse have some time behavior, E(t). We
can, of course, consider the Fourier transform of this,
and its inverse (I’ll drop vectors to simplify the nota-
tion):
E(ω) =1
2π
∫ ∞
−∞E(t)eiωtdt
⇔ E(t) =
∫ ∞
−∞E(ω)e−iωtdω
(4.2)
Two FT facts will be useful. First, because E(t) is real,
we know that
E(−ω) = E∗(ω) ; |E(−ω)|2 = |E(ω)|2 (4.3)
(that is, the negative frequencies contain no new infor-
mation). Second, a general result from Fourier trans-
forms (Parseval’s theorem) tells us that∫ ∞
−∞E(t)2dt = 2π
∫ ∞
−∞|E(ω)|2dω
= 4π
∫ ∞
0|E(ω)|2dω
(4.4)
(I’ve used 4.3 in the last step).
Now: the total energy per unit area emitted in the radi-
ation pulse is
dW
dA=
c
4π
∫ ∞
−∞E(t)2dt (4.5)
(to see this, start with the Poynting flux, and remember
that E = B for an EM wave in cgs). Now by (4.4), we
can rewrite this as
dW
dA= c
∫ ∞
0|E(ω)|2dω (4.6)
20 Physics 426 Notes Spring 2016
So: we’re just about there. What we do, essentially,
is think of the integral in (4.6) as an integral over
the frequency spectrum of the radiation pulse: that is,
c|E(ω)|2 measures the energy in the pulse “at ω”.3
Thus: looking back to (4.1), we can connect this for-
malism to the Larmor result. Compare (4.1) to (4.6):
they both described the total energy within the pulse.
So, think about the Fourier transform of some compo-
nent (x, y or z) of the acceleration:
ai(t) =
∫ ∞
−∞ai(ω)e
−iωtdω (4.7)
We can thus identify
P (ω) =8π
3
e2
c3|a(ω)|2 (4.8)
with what we want, namely, the spectrum of the radi-
ation. The 4π difference between the frequency-space
definition here and the real-time definition (4.1) arises
in the Fourier transform (4.4).
4.2 Bremsstrahlung I: single particle
We can now apply this to radiation from an electron-
ion collision, using the usual geometry (which you re-
member from chapter 3 of our Phys 425 notes).
b
v
Figure 4.2 The usual geometry, as an electron is deflected
by an ion. The impact parameter is b; let the electron move
in the x direction to start, with velocity v.
The two components of acceleration are
ax =e2vt
m(b2 + v2t2)3/2
az =e2b
m(b2 + v2t2)3/2
(4.9)
where we have used the impact parameter, b, have set
the origin of time at the time of closest approach, and
3There is an important technical detail here. The expression
(energy/area) in (4.5) or (4.6) is not per unit time, rather it’s inte-
grated over the pulse. If we tried to do this argument “per dt” and
“per dω”, we’d violate the uncertainty relation between ω and t in
the wave packet. However, if the pulses repeat frequency, one can
formally take limits and get the same result ... cf. RL for details
here.
have assumed the particle suffers only a small deflec-
tion, so that the v does not change by much. The ra-
diated spectrum will depend on the FT of this accel-
eration. Without doing this out algebraically, we can
predict the answer, using what we know about Fourier
transforms. In particular, we note that
(i) The z-component of the acceleration will be the
dominant factor over the course of the encounter be-
cause it does not go to zero at closest approach.
(ii) Since a(t) is large only when the two particles are
close together – just as we argued in the Coulomb colli-
sion discussion – P (t) will be significant only for times<∼ 2b/v.
(iii) Therefore, we expect the spectrum will be domi-
nated by the FT of az , and that the FT will have power
at frequencies ω <∼ v/2b.
Doing the actual work, Longair gives the result for both
parallel and perpendicular acceleration:
ax(ω) =1
2π
∫ ∞
−∞
e2vt
me
eiωt
(b2 + v2t2)3/2dt
=1
2π
e2
mebv
2ωb
viK0
(
ωb
v
)(4.10)
and
az(ω) =1
2π
∫ ∞
−∞
e2b
me
eiωt
(b2 + v2t2)3/2dt
=1
2π
e2
mebv
2ωb
vK1
(
ωb
v
)(4.11)
where Ko(ωb/v) and K1(ωb/v) are modified Bessel
functions. We can find analytic forms for K1 and Ko
in the limits of large and small arguments:
K0(x) → − lnx x ≪ 1
→√
π
2xe−x x ≫ 1
(4.12)
and
K1(x) →1
xx ≪ 1
→√
π
2xe−x x ≫ 1
(4.13)
From this, we can add both acceleration terms
(squared) to get the radiated spectrum:
P (ω) =8πe2
3c3[
|ax(ω)|2 + |az(ω)|2]
(4.14)
Physics 426 Notes Spring 2016 21
Using the analytic expressions for the modified Bessel
functions, we find the limiting forms for the spectrum
radiated in a single-particle encounter:
P (ω) ≃ 8
3π
e6
m2ec
3v2b2ω ≪ v
2b
P (ω) ≃ 8
3π
e6
m2ec
3v2b2e−ωb/v ω ≫ v
2b
(4.15)
You should note that the exponential cutoff in the sec-
ond equation, here, means there is effectively no radi-
ation above ω ∼ v/b — which is consistent with what
we expect from the duration of the wave packet.
The low frequency limit of (4.15) is worth commenting
on. P (ω) → constant as ω → 0, and is well behaved.
However, the photon emissivity, P (ω)/hω diverges as
1/ω as ω → 0. This is a well-known problem in both
classical and quantum electrodynamics, known as the
“infrared divergence.” This is essentially an artifact of
our derivations, rather than a problem with the physics
(for instance, see Jauch and Rohrlich, The Theory of
Photons and Electrons). Here, we will simply note that
the energy lost is finite, and that self-absorption in any
finite system (see below) will keep us from ever seeing
this divergence anyway. We will, therefore, carry on
happily.
4.3 Bremsstrahlung II: from a plasma
Now, we want to extend this to consider a particle in
a plasma, and to take all of its collisions into account.
We did this last term, when we derived Coulomb scat-
tering – we had to take all impact parameters into ac-
count. We’ll do essentially the same thing here.
First, consider the range of impact parameters that one
particle encounters. Since the number of ions that one
electron, at velocity v, sees per second at impact pa-
rameter b is 2πnivbdb, we can find the total radiated
spectrum from that electron,
P (ω, v) =
∫ bmax
bmin
P (ω, v, b)2πnivb db
=
∫ bmax
bmin
8
3π
e6
m2eb
2v2c3niv2πb db
=16
3
e6ni
m2evc
3ln
(
bmax
bmin
)
(4.16)
Again, the range of impact parameters must be chosen
with some physics in mind. And again, luckily, our
choice only affects the answer logarithmically. Typical
choices are bmin ∼ e2/mev2, and bmax ∼ v/ω or
∼ h/2mev.
Next, we use this to find the total energy loss rate for
one particle. We do this by integrating P (ω, v) over
all frequencies. Since P (ω, v) is only a weak function
of ω (it appears only in ln(bmax/bmin)), up to the fre-
quency cutoff ωmax ≃ mev2/h (which is the highest
photon frequency we can expect, from simple energy
conservation), we have
P (v) =
∫ ωmax
ωmin
P (ω, v)dω
=16
3
e6
c3mehln
(
bmax
bmin
)
niv
(4.17)
This has the functional form P (E) ∝ E1/2.
Returning to the single particle spectrum (4.16), we
can now integrate over all particles in the plasma to get
the total emissivity from that plasma. We need to know
the distribution of electron speeds, and we will assume
a Maxwell-Boltzmann distribution. We also switch
from ω to ν = ω/2π, to connect with observations;
and we derive jff (ν), the emissivity per steradian, to
connect with the radiative transfer applications, above.
(That means simply a 4π factor, since the single parti-
cle emission is essentially isotropic). Thus,
jff (ν) =1
4π
∫ ∞
0P (ω, v)f(v)dv (4.18)
with f(v) assumed to be the Maxwellian of (3.9), nor-
malized to ne. This gives us
jff (ν) =8
3
(
2π
3
)1/2 e6
m3/2e c3
× neni
(kBT )1/2gff (ν, T )e
−hν/kBT
(4.19)
Numerically, with everything in cgs units, this is
jff (ν) =5.44 × 10−39gff (ν, T )neni
T 1/2e−hν/kBT
erg s−1cm−3Hz−1str−1 (4.20)
In this expression, we have implicitly defined the
Gaunt factor, gff (ν, T ). It arises from the velocity
dependence of ln(bmax/bmin), inside the integral in
(4.18): the essential part of the integral is∫
1
vf(v) ln
(
bmax(v)
bmin(v)
)
→ 〈1v〉〈ln
(
bmax(v)
bmin(v)
)
〉(4.21)
22 Physics 426 Notes Spring 2016
We see that the 〈1/v〉 becomes the (kBT )−1/2 term;
the mean of the logarithmic factor becomes gff (ν, t),the Gaunt factor. Note that both bmax and bmin might
be functions of ν and of v. As with Coulomb scatter-
ing, different expressions, corresponding to different
choices of bmax and/or bmin, are used in different sit-
uations. A couple of common cases are, first, in the
radio range, with hν ≪ kBT :
gff (ν, T ) ≃√3
πln
(
2
π
(kBT )3/2
e2m1/2e ν
)
≃ 10
(
1.0 + 0.1 logT 3/2
ν
) (4.22)
Second, in the X-ray range, with hν <∼kBT , people use
gff (ν, T ) ≃√3
πln
(
kBT
hν
)
(4.23)
Finally, the total emissivity of the plasma can be found,
by integrating jff (ν) over all frequencies and all solid
angles. This is
εff = 4π
∫ ∞
0jff (ν)dν
=
(
2πkB3me
)1/2 32πe6
3hmec3neni〈gff 〉T 1/2
≃ 1.4 × 10−27neni〈gff 〉T 1/2 erg cm−3s−1
(4.24)
where 〈gff 〉 is the mean Gaunt factor, averaged over
frequency.
We are also interested in free-free absorption. This
is the inverse of the emission process; a free elec-
tron absorbs a photon (the ion must be there, as well,
to conserve momentum and energy at the same time).
For absorption by a Maxwellian plasma, for which we
have just derived the emissivity jff (ν), we can get the
absorption coefficient, κff (ν), immediately from Kir-
choff’s law (3.34):
κff (ν) =4
3π
(
2π
3
)1/2 e6nenigff (ν, T )
m3/2e c(kBT )3/2ν2
≃ 0.018gff (ν, T )neni
T 3/2ν2cm−1
(4.25)
This expression is valid in the Rayleigh-Jeans limit
(3.22); and the second expression is all in cgs units).
You should note that this expression also contains the
Gaunt factor. The most common use of free-free ab-
sorption is in the radio range (given the 1/ν2 form),
and a commonly used form of the absorption, which
includes a particular expression for the Gaunt factor, is
κff (ν) ≃ 0.08235neni
T 1.35ν2.1GHz
pc−1 (4.26)
Note the oddball units: ν is in GHz; κ is in inverse
pc(!); but ne, ni and T are still in cgs.4 Avrett, Fron-
tiers of Astrophysics, says this is good for 0.1 < ν <50 GHz, and for 6000 < T < 18, 000 K (which de-
scribes HII regions and much of the warm, ionized
ISM, for instance).
-1 0 1 2-3
-2.5
-2
-1.5
log frequency
Figure 4.3 Illustrating the full bremsstrahlung spectrum
we expect from a source which has a finite size, and thus
a finite optical depth. Note, the jagged appearance of the
high-frequency exponential is the fault of my simple plot-
ting program, not the physics.
Finally: let’s connect this to bremsstrahlung emission
from a finite object. Think, for instance, about a galac-
tic HII region (such as the Orion nebula). Typical tem-
peratures are T ≃ 104 K, and typical densities might
be n ∼ 100 cm−3; the size might be on the order of
a pc. We can estimate τff (ν) = κff (ν)ℓ, where ℓ is
again the line-of-sight thickness of the object, and we
find τff (ν) ≃ 1 for frequencies in the low radio range
(a fraction of a GHz, say). Below this frequency the
source will be optically thick, and the emergent inten-
sity will obey Iν = Bν(T ) ∝ ν2, from the Rayleigh-
Jeans limit of the Planck function. At higher frequen-
cies, the source is optically thin, and the emergent in-
tensity has the same frequency dependence as the fun-
damental emissivity: Iν ∝ jff (ν, T ). Thus, the inten-
sity will be approximately constant at higher frequen-
cies. At very high frequencies, hν ∼ kBT (that is,
tens of eV → 1016 Hz or so), the exponential cutoff
will appear. Figure 4.3 sketches this behavior.
4That’s astronomers for you ..
Physics 426 Notes Spring 2016 23
References
I’m mostly following a fairly nice discussion given in
• Longair (High Energy Astrophysics, Vol II);
and pulling some of the discussion on foundations
from
• Rybicki & Lightman, Radiative Processes in Astro-
physics
Key points (for chapter 4)
• Total power (Larmor formula);
• Wave content ↔ radiation spectrum; “seat-of-the-
pants” FT analysis.
• Bremsstrahlung: basic physical picture, single parti-
cle
• Bremsstrahlung: from a plasma; intrinsic spectrum
and power
• Bremsstrahlung: emissivity, absorption coefficients,
total spectrum
24 Physics 426 Notes Spring 2016
5 Thermal state of the ISM
What determines the energy balance of diffuse astro-
physical gases? In general, systems which have had
time to reach a thermal balance will have a tempera-
ture determined by the balance of heating and cooling
rates. For most phases of the ISM (although probably
not including the hot, coronal gas), the cooling is due
to radiation, from an optically thin gas. The specific
radiation mechanisms which cool a cloud or nebula
will depend on the composition, the internal excitation
states and the temperature of the cloud. This part of
the problem is fairly well understood by now, at least
for the low-density conditions which describe the ISM.
The heating mechanism must also be specified; in gen-
eral, the ISM can be heated by the local radiation field,
and by the local cosmic ray population. Since the rate
of energy transfer from either of these mechanisms can
depend on the local ionized fraction, we must also con-
sider ionization balance. In this chapter, we will first
look at some general features of cooling and heating in
this context. We will then look at two well-understood
examples, one being the thermal state of an HII region,
and the other being the multiphase ISM.
NOTE to the student: This chapter contains
quite a few details and a lot of unfriendly-
looking equations. That’s because we need
these details in order to understand the vari-
ous processes whose balance determines the
thermal state of the ISM. In addition to the
Key Points at the end of the chapter, I’m
adding “Look ahead” comments to each of
the major sections .
5.1 Heating and cooling: general considerations
The energetics of astrophysical fluids or plasmas can
be more complicated than lab fluids, because direct
heating and cooling can be important. By “direct” I
mean that a small volume element, deep inside a cloud,
can exchange energy directly with the outside world
via photons or other particles.
LOOKING AHEAD: The biggest result here
is the general cooling curve, Λ(T ), shown in
Figure 5.1, and described in the text. The
other important bit is the list of processes
which contribute to heating the ISM.
5.1.1 Cooling
For energy loss, the most important astrophysical case
is radiation. If the plasma or cloud is optically thin,
radiation generated inside the volume element escapes
the cloud without further interaction. The net energy
loss (per volume per time, frequency-integrated) is of-
ten called
Λ(n, T ) = n2L(T ) (5.1)
and has dimensions, erg cm−3 s−1. Note the sepa-
ration into n2 and a function L(T ) of T only. It re-
flects the fact that radiative cooling processes are all
two-body processes (electron-ion scattering, electron-
atom collisions, etc). The total cooling rate from a ther-
mal astrophysical plasma is due to the sum of all possi-
ble spectral lines emitted and also continuous emission
due to electron-ion collisions. You’ve already seen the
latter; it’s bremsstrahlung. The former can be quite
complicated – one has to know the excitation state of
each level of each atom (which depends on the tem-
perature of the gas and the microphysics of excitation
and de-exciation processes) – and must be calculated
numerically.
Figure 5.1 Cooling rates as a function of temperature, for
diffuse astrophysical plasma at a density of 1 cm−3. Differ-
ent curves indicate differing heavy metal abundances (rela-
tive to hydrogen) by number. Cooling rates for other densi-
ties can be obtained by scaling as n2. From Maio et al 2007,
MNRAS 379, 963; see also an older version in Spitzer Fig-
ure 6.2.
Physics 426 Notes Spring 2016 25
Figure 5.1 shows the result, from a calculation (orig-
inally due to Raymond, Cox & Smith 1977), which
adds up all of the different cooling mechanisms, to get
what’s called the standard cooling curve. The main
coolants, in the various temperature regions, are:
• T < 104 K: collisional excitation of fine structure
levels of various heavy elements; lines mostly in the
IR.
• T ∼ 104 K: excitation of hydrogen lines; the
great abundance of hydrogen makes this the dominant
coolant in the temperature range where H is signifi-
cantly excited at or above the first excited state.
• T ∼ 104 − 107 K: excitation of optical/UV lines of
heavy elements, often forbidden lines.
• T >∼ 107 K: no species are left un-ionized, so
bremsstrahlung is the only coolant.
At temperatures T >∼ 109 K, other reactions – such as
pair production – become important; very few astro-
physical plasmas are this hot, and if they are, other
physics is probably important as well.
5.1.2 Heating
Direct heating of astrophysical plasmas isn’t so simple,
because it depends on the environment of the cloud,
not just its internal state (density and temperature). We
know, in general, the heating sources for the diffuse
ISM. Stars are the fundamental energy sources; they
lose energy, and supply power to the ISM, by starlight
and also through dynamic input from such events as su-
pernovae and stellar winds. What we need to quantify
here, however, is the microphysics: how does this en-
ergy couple to some small test volume deep within an
interstellar cloud? This area is still under discussion.
Here I lean on the discussion from McKee (1995),
which focuses on the diffuse ISM; but add photoion-
ization heating which is important for HII regions.
• Cosmic Rays. These high-energy particles pen-
etrate the cold ISM, and can transfer energy both
through ionization of the neutral component and
through Coulomb interactions with the ionized com-
ponent. This is the heating mechanism first suggested
by Field, Goldsmith, &Habing (1969), in their origi-
nal model of the multiphase ISM. Since then other au-
thors have argued that cosmic rays can’t supply enough
power, and that other possibilities are more important.
• X-rays. These have also been suggested, as they can
also penetrate the ISM. They heat by ionization. How-
ever, most authors seem to agree that their heating rate
is also too low to be useful.
• Magnetic dissipation. This is attractive but harder
to quantify. The idea is that the large-scale galactic
field will dissipate locally, either by local reconnec-
tion events (similar to solar flares), or by dissipation of
MHD waves. The latter in particular could be a heating
mechanism for the cool ISM, as the waves could prop-
agate fairly far from their sources before dissipating.
I have not seen this treated quantitatively, however, in
any ISM modelling.
• Photoelectric heating. This seems to be the cur-
rent favorite. Starlight is abundant in the galaxy (it
also has a density ∼ 1 eV/cm3); can it couple to the
ISM? Most starlight is too cool to ionize the ISM. It
can, however, be absorbed by interstellar grains which
then eject electrons, in a standard photoelectric effect.
This can be quantitatively effective if very small grains
and also large organic molecules (polycyclic aromatic
hydrocarbons, PAH’s) are included in the models.
• Photoionization heating. This will be treated in
more detail immediately below. It is unlikely to be im-
portant in the diffuse ISM, because there are not very
many “loose photons” with hν > 13.6 eV; but it is the
dominant heating mechanism close to hot, young stars
(and probably close to active galactic nuclei).
For our purposes here, the total heating rate – whatever
it may be – will be called
Γ = nH (5.2)
and has units erg cm−3 s−1. Once again, we’ve ex-
tracted the density dependence as most of the heating
mechanisms described here depend on the first power
of the density (i.e., on how many atoms are around to
be heated).
5.2 HII regions
Let’s start with a spherical chicken. That is, we will
start with the ionization and thermal structure of a sim-
ple, one-star HII1 region. Consider one hot star (O or
B type, with a significant part of its luminosity above
the ionization edge of hydrogen, 13.6 eV), sitting in
the ISM. The UV photons from this star will ionize
and heat the local hydrogen, producing an HII region
1Notation: HII (called “H-two”) refers to ionized hydrogen; HI
(“H-one”) is neutral hydrogen. And just to be confusing, H2 (also
called “H-two”) is molecular hydrogen.
26 Physics 426 Notes Spring 2016
around the star. Within an HII region, the central star
(or stars) will dominate the energy and ionization bal-
ances, which simplifies the problem (compared to the
more general case of the diffuse ISM). To start, then,
we will therefore consider the equilibrium of a purely
photonionized nebula.
LOOKING AHEAD: The important features
in here are (i) the definitions of ionization
and recombination rates (say equations 5.4
and 5.5); their use in ionization balance
(equations 5.7, 5.8); the Stromgren sphere
(5.17); the general discussion of heating and
cooling (§5.2.2); and the final result, 8000-
10,000 K “always”.
5.2.1 Ionization structure
Let the central star have a spectrum L(ν) (in erg/s-Hz).
At a distance r from the star, the mean intensity is
J(ν, r) =L(ν)
4πr2e−τ(ν,r). (5.3)
Look back to chapter 3: J(ν, r) = 14π
∫
I(ν, r)dΩ is
the mean intensity averaged over solid angle. The ex-
pression in (5.3) describes the intensity from the cen-
tral star, attenuated by some optical depth τ(ν, r) to
be evaluated below. Photons above hν1 = 13.6eVwill ionize hydrogen; the ionization cross section is
σion(ν) ≃ 6.6× 10−18(ν/ν1)−3 cm−2. We can there-
fore write down the ionization rate per hydrogen atom,
ϕuv(r) =
∫ ∞
ν1
4πJ(ν, r)
hνσion(ν) dν (5.4)
Note that ϕuv has units s−1; the ionization rate per
volume is nHIϕuv, units cm−3 s−1.
We also need the recombination rate to level j, from
the continuum; call it neαj(T ) recombinations per sec-
ond per atom. From the recombination cross section,
again assuming a Maxwellian distribution of free elec-
tron velocities, αj(T ) is
neαj(T ) =
∫
vσrec,jf(v)dv
≃ 2.1× 10−11gjne
[1 + j2kBT/13.6eV)] j2T 1.2
(5.5)
where gj is the statistical weight of that level. This can
be summed over all levels to get the total recombina-
tion rate per atom,
α(T ) =∑
j
αj(T ) ≃ 2.1 × 10−11Φ(T )
T 1/2cm3s−1
(5.6)
where Φ(T ) is yet another order-unity factor, which
varies only slowly with temperature in the range 102−105K.
In equilibrium, the ionization rate will be balanced by
the recombination rate, nHIIneα(T ) (recombinations
per volume per second). The balance is then given by
nHIϕuv = nHIIneα(T ) (5.7)
Now, the total hydrogen density is n = nHI +nHII ; if
the nebula is mostly hydrogen, ne ≃ nHII . Thus, we
can define x = nHII/n as the ionized fraction (so that
n(1− x) is the neutral fraction), and write (5.7) as
1− x
x2=
nα(T )
ϕuv(5.8)
We note, ϕuv is a function of position in this case, from
(5.4).
Consider the solutions to (5.8) for x. We know 0 ≤x ≤ 1, by definition. Now, close to the star, the right
hand side of (5.8) will be ≪ 1. We therefore know
x ≃ 1, and the quadratic solution of (5.8) will be
1− x ≃ nα(T )
ϕuv≪ 1 (5.9)
Far from the star, the right hand side will be ≫ 1 (as
ϕuv drops), and the solution will be
x ≃(
ϕuv
nα(T )
)1/2
≪ 1 (5.10)
In addition, the region (of r) over which nα(T )/ϕuv
changes from large to small turns out to be small (you
can show that this region has width ∆r ∼ the mean
free path of a photon in the neutral gas). Thus, we
expect the ionization structure to go from fully ionized,
to hardly ionized, over a very short distance.
NOW, We want to study the structure of the nebula in
a bit more detail, and to find the location of the ion-
ized/neutral transition. To start, we need the optical
depth, from the star out to radius r:
τ(ν, r) =
∫ r
0n(1− x)σion(ν)dr
≃ n(1− x)σion(ν)r
(5.11)
Physics 426 Notes Spring 2016 27
where the last step assumes a uniform density and that
(1 − x) ≃ 0 ≃ constant (which is true within the ion-
ized region). Now, consider the local radiation density,
u(ν, r) = J(ν, r)/c. Using (5.3), we can write
d
dr
(
4πr2cu(ν, r))
=d
dr
(
L(ν)e−τ(ν,r))
= −L(ν)e−τ(ν,r)dτ(ν, r)
dr
(5.12)
And, using (5.11), this becomes
d
dr
(
4πr2cu(ν, r))
= −L(ν)e−τ(ν,r)n(1− x)σion(ν) .
(5.13)
Now, we can (a) use (5.7) to eliminate n(1− x) on the
right hand side; (b) multiply both sides by dν/nν, and
(c) integrate over ν:
d
dr
[
4πr2c
∫ ∞
ν1
u(ν, r)
nνdν
]
= −4πr2cα(T )n2x2
ϕuv
∫ ∞
ν1
u(ν, r)σion(ν)
hνdν
(5.14)
At this point, we can identify terms. The quantity in
brackets on the left of the equation is the number of UV
photons per second crossing the surface at r, Suv(r).The integral on the right side is just ϕuv(r)/c. Thus,
the equation simplifies quite nicely to
dSuv(r)
dr= −4πr2α(T )n2x2 (5.15)
For a constant density region, in which x ≃ 1, we can
easily solve this equation:
Suv(r) = Suv(0) −4π
3r3α(T )n2x2 (5.16)
where Suv(0) is the photon flux at the star.
From this, we can define the Stromgren radius as the
distance at which the photon flux goes to zero:
Suv(0) =4π
3R3
sα(T )n2x2 (5.17)
This limit defines the Stromgren sphere. We note that –
due to the assumption of local ionization balance – the
expression for Rs only depends on the total number of
UV photons, not on their energy, or how many central
stars there are, or the ionization cross section, or any
number of physical parameters one might have thought
interesting. Rs is determined only by Suv(0)/n2. (Re-
member that x ≃ 1 inside Rs). The physical picture
is, simply, that the size of the HII region – Rs – is set
by the volume inside of which the number of recom-
binations per second exactly balanced the number of
ionizing photons put out by the star per second.
5.2.2 Energy balance and temperature
Photoionization also provides the heating for the neb-
ula. A photon of energy ν > ν1 ionizes an atom, and
the leftover energy h(ν − ν1) goes to kinetic energy of
the free electron. This electron can then share the en-
ergy with the ions, through Coulomb collisions, so that
this becomes a general heating mechanism for the gas.
The heating rate per HI atom can be written
∫ ∞
ν1
4πJ(ν)
hνσion(ν)h(ν − ν1)dν
≃ ϕuv〈h(ν − ν1)〉(5.18)
where the second expression defines the mean energy
transfer per ionization (with the mean taken over the
input photon spectrum). Note, we have suppressed the
r dependence in J(ν), to ease the notation. Thus, the
heating rate per volume is
Γuv = nHIϕuv〈h(ν − ν1)〉 (5.19)
We can again assume ionization balance, and this can
be written in terms of the local density and tempera-
ture, as
Γuv = n2x2α(T )〈h(ν − ν1)〉 (5.20)
To determine the equilibrium temperature, we want to
balance Γuv against all of the important radiative cool-
ing mechanisms (assuming the HII region is optically
thin, which is a good assumption for visible ones – the
ones in the pretty pictures). It turns out that there are
two important types of coolants – recombination lines
from hydrogen (and helium, which is a small correc-
tion), and collisionally excited lines from heavy ele-
ments. To do the calculation, we have to to compute
the cooling rates numerically for these lines (or find
someone who has done it already!).
The hydrogen recombination cooling rate is
Λrec = nHII
∑
j
∫
σrec,j(v)vf(v)1
2mev
2dv (5.21)
28 Physics 426 Notes Spring 2016
Now, the integral-sum on the RHS is essentially the net
recombination rate, (cf.equation 5.5), times the mean
energy released per recombination:
Λrec ≃ n2x2α(T )kBT . (5.22)
The integral is over the electron distribution function,
and the sum is over all relevant energy levels.
Can hydrogen cooling alone account for the temper-
ature of an HII region? We observe temperatures<∼ 104 K. Now, if the heating is by hydrogen ioniza-
tion, and the cooling is by hydrogen recombination,
our thermal balance would be
Γuv(T ) = ΛH,rec(T ) (5.23)
Comparing (5.20) and (5.22), we see that this solves to
kBT ≃ 〈h(ν − ν1)〉 (5.24)
(everything about the density and ionization state drops
out, note – due to the ionization rate balance). But
this is too hot; recall hν1 = 13.6 eV. Thus, simple
estimates don’t work here – we need to include heavy
element cooling, which turns out to be stronger, and
results in the lower temperatures we see.
For the heavy element cooling rates, we can write
schematically (for element X)
ΛX =∑
j,k
nX,jAX,jkhνjk (5.25)
where the sum is over all “upper” and “lower” states,
and the level populations nX,j must be determined by
the methods discussed previously. An important fact to
know is that the levels are collisionally excited (but ra-
diatively de-excited, of course). Thus, the net cooling
rate Λx ∝ nenX ∝ n2, since it is a collisional process.
Thermal balance in the nebula can then be written as
Γuv(T ) = Λrec(T ) +∑
X
ΛX(T ) . (5.26)
Solutions of this equation determine the temperature of
the nebula. From (5.20), and the discussion just above,
we see that both sides ∝ n2, so that the equilibrium
temperature does not depend on the density. The tem-
perature can be found numerically.
This is the big result: the temperature does not depend
on the density; only on the details of the ionizing spec-
trum and coolants. Thus, the temperature of any (pho-
toionized) HII region is ∼ 8000−10, 000 K. This result
is independent of the density (nearly) and of the num-
ber of stars ionizing the region; it has some dependence
on the ionizing spectrum and the abundance of heavy
elements, primarily oxygen. Through the level popula-
tions it does have a weak dependence on the density.
5.3 The diffuse ISM: multiphase equilibrium
For the general diffuse ISM, things aren’t quite so sim-
ple. However, Field etal (1969) developed a very nice
semi-analytic formulation; here I simplify the algebra2
by extracting only the dominant terms. We begin by
determining what the most important cooling and heat-
ing mechanisms are. We start with cooling, which is
the simpler of the two, as it only depends on micro-
phyics.
LOOKING AHEAD: The big results here are
(i) the general idea of setting up a “heating
= cooling” balance for the ISM; and (ii) the
result, that the cold and warm phases which
coexist in the ISM are thought to be two sta-
ble solutions of a “heating = cooling” bal-
ance.
5.3.1 Cooling: what dominates here?
The ISM cools by radiation (it’s optically thin to
many/most photons). The radiation is generated, typi-
cally, by collisional excitation of some atom or other,
by a free electron. The atom de-excites by radiation,
leading to a net energy loss from the system. If we
want to find which of all possible line transitions are
important for cooling, we can think of several criteria.
• The species must be fairly abundant;
• its excitation energy must be comparable to, or less
than, the typical electron thermal energy;
• it must have a large cross section for such excitation;
• it must have a high probability to de-excite before
another collision occurs;
• and it must generate photons to which the ISM is
optically thin.
Given this general argument, we find that two of the
most important coolants for the temperature range that
describes the cold and warm hydrogen (which is a few
×102 to a few ×103 K) are particular lines from hydro-
gen and from carbon. We can approximate the cool-
2honestly, folks!
Physics 426 Notes Spring 2016 29
ing rate of these by saying that the line radiatively de-
excites as soon as it is excited; thus, the cooling rate is
just proportional to the excitation rate. But the colli-
sional excitation rate is, typically,
qX;ij ≃ qXoT−1/2e−hνij/kBT (5.27)
for excitation from level i to level j of species X. The
term qXo is a numerical factor containing atomic con-
stants.3 For our two main coolants, the cooling rate can
be written,
ΛH ≃ nenHIqH
T 1/2e−hνH/kBThνH (5.28)
and
ΛC ≃ nenHInC
nH
qC
T 1/2e−hνC/kBThνC (5.29)
Here, we are assuming that x ≪ 1, so that nHI ≃n. We note that the line frequencies are given by
hνC/kB ≃ 92 K, and hνH/kB ≃ 105 K. Thus, at
low temperatures C cooling dominates, and at higher
temperatures H cooling dominates.
5.3.2 Heating: by cosmic rays?
As an example of how this can be quantified, consider
cosmic ray heating. This may not be the dominant
heating; however it is easy to write down analytically
(as Field etal did), and therefore provides a useful ex-
ample of the analysis. In addition, cosmic ray heat-
ing is proportional to the local density, just as X-ray
heating and photoelectric heating are (and probably as
magnetic heating would be, too). Thus, the analysis
I present here could be extended to PAH heating by
changing the constants; the results for the multiphase
ISM should be fairly robust.
In cosmic ray heating, the neutral fraction will be
heated by ionization. If σion(E) is the cross section
for ionization by a cosmic ray of energy E, fcr(E) is
the density of cosmic rays at energy E, and ∆E is the
excess energy the electron comes away with in this ion-
ization, the ionization rate per volume can be written,
nHIϕcr = n(1− x)
∫
fcr(E)v(E)σion(E)dE
(5.30)
3The exponential, and the inverse T 1/2, arise from an integral
of the collision rate, 1/nσ(v)v, integrated over the electron veloc-
ity distribution; see if you can justify this for yourself.
From this, the heating rate due to cosmic ray ionization
can be written,
Γcr,ion = n(1− x)
∫
fcr(E)v(E)σion(E)∆EdE
= n(1− x)ϕcr〈∆E〉(5.31)
in direct analogy to (5.20). Numerically, this turns out
to be
Γcr,ion ≃ 5× 10−12ϕcrn(1− x) erg cm−3 s−1
Current estimates for ϕcr, averaged over the disk, sug-
gest ϕcr ≃ 10−17 s−1.
The ionized fraction of the gas will be heated by
Coulomb collisions with the cosmic rays. A single cos-
mic ray, once it is relativistic, loses energy as a rate
PC(E) ≃ 6×10−19nx erg/s to the ionized component
(this number comes from the Coulomb collision rate
for relativistic particles). Thus,
Γcr,C =
∫
fcr(E)PC (E)dE (5.32)
To simplify the algebra below, we note that Γcr,C can
be related to ϕcr, since both involve an integral over
fcr(E), the cosmic ray distribution function. We will
write this relation as Γcr,C = Aϕcrnx, where A con-
tains a mean of PC(E)/σion(E), and other (numerical)
factors. Finally, we can also write down the ionization
state of the gas, due to the cosmic rays:
ϕcrn(1− x) = n2x2α(T ) (5.33)
which has limiting solutions for x, just as in (5.7).
5.3.3 Thermal balance and multiphase
equilibrium
We will follow the method first presented by Field et al
(1969), and will approximate the behavior of the heat-
ing and cooling functions to allow us to find an analytic
solution for the equilibrium temperatures. This anal-
ysis addresses the two cooler phases of the ISM, the
cold and warm HI distributions. (The hot, coronal gas
is probably heated dynamically, by supernova shocks
and/or stellar winds, and also cooled dynamically by
its expansion out of the galactic plane; it will not be
included in this analysis).
Now, the general thermal balance equation, for heating
by cosmic rays and cooling by these two spectral lines,
30 Physics 426 Notes Spring 2016
from C and H, is
ϕcrn(1− x)〈∆E〉+ ϕcrAnx = n2x(1− x)
×[
qH
T 1/2e−hνH/kBThνH +
nC
nH
qC
T 1/2e−hνC/kBThνC
]
(5.34)
This equation gives the implicit (n, T ) solution, for
the density-temperature relation of the equilibrium gas.
We show this solution, below; but it seems worth get-
ting some insight into its behavior first. Consider only
the low-temperature regime, where we can ignore H
cooling, and also ignore Γcr,C (since x ≪ 1). The
balance is then, approximately,
nϕcr(1− x)〈∆E〉≃ n2x(1− x)
nC
nH
qC
T 1/2e−hνC/kBThνC
(5.35)
But from this, using the x ≪ 1 limit of (5.33), we find
that
n ∝ Tα(T )e2hνC/kBT ∝ T 1/2e2hνC/kBT (5.36)
This gives us an approximate analytic version of the
(n, T ) relationship for T ∼ 102 − 103 K. We can find
a similar expression for the high-T regions, where H
cooling dominates. The net solution looks like the car-
toon in Figure 5.2. This solution can also be turned
into a (p, T ) solution, using p ∝ nT , also in Figure
5.2.
But this is, then, (almost) the end of the analysis. That
is, since we know the pressure of the system – as we
do, observationally (nT ∼ few × 103 cm−3 K), we
can find the (p, T ) solutions which are allowed by this
thermal balance. Figure 5.2 illustrates the solutions;
in principle anywhere on the (n, T ) or (p, T ) loci is
a solution. We restrict the possibilities by specifying
the pressure – equating it to the ISM pressure (which
is known). This gives us three possible solutions (as in
Figure 5.2). But will all three exist in the ISM?
5.3.4 Is the thermal balance solution stable?
The final step of the argument is to argue, as Field etal
did, that the outer two solutions – T ∼ 102 K and
T ∼ few × 103 K – are the two stable HI phases; and
that the middle T solution is unstable. The argument
goes as follows. Consider one particular equilibrium
state, with density and temperature specified, as well
as with external pressure fixed. For instance, look the
log n
log T
CII
HII
p
CIIlog (nT)
HII
log T
ISM
Figure 5.2 Top; sketch of the (n, T ) solution from (5.34).
The lower temperatures are controlled by CII cooling; the
higher temperatures by HII cooling. Any (n, T ) pair on this
curve is a possible solution; but not all will exist in the ISM.
Bottom: the same solution, but now plotting “pressure” nTagainst T . The dotted line indicates the ISM pressure, which
picks out (n, T ) solutions that might exist in the ISM. The
small circles show three possible solutions; however only
the two outer ones are thermally stable. When we put in
real numbers, these stable solutions correspond to the cold
and warm phases of the ISM.
middle T solution in Figure 5.2. Consider a slight com-
pression of this state, in which the temperature drops a
little bit (this is required from the (n, T ) solution in
this region). But the pressure must also drop in the
perturbation (from the (p, T ) curve); so the larger ex-
ternal pressure will compress the perturbation still fur-
ther – and the density will drop still more, and it will
cool faster . . . and so on. That is, this system is
thermally unstable; a slight compression will collapse
and cool, and a slight rarefaction will expand and reach
higher temperatures. The timescale for this runaway
∼ tcool ∼ kBT/Λnet(T ), as always; for the ISM, this
is a short time, so the instability will develop rapidly.
Conversely, for the outer two solutions, the slope of
the (p, T ) curve is reversed, so that – say, a small com-
pression will reach higher pressures, and expand back
to its original state (or vice-versa for a small rarefac-
tion). Thus, the two outer states should be stable, and
Physics 426 Notes Spring 2016 31
are believed to represent the two cool phases of the
ISM.
More formally, Field etal (1969) worked with the net
cooling function, C = Λ − Γ = n2L − nH. Again,
the cooling term is Λ = n2L as in Figure 5.1. The
heating term is written as Γ = nH to highlight the
linear dependence on density n. They showed that the
system is unstable if
∂C∂T
∣
∣
∣
p=
∂C∂T
∣
∣
∣
n− n
T
∂C∂T
∣
∣
∣
T< 0 (5.37)
Now, using C = 0 (which means we start in thermal
balance, before we perturb the system) in (5.37) turns
this condition into
∂C∂T
∣
∣
∣
p= n2 dL
dT− n2L
T< 0 (5.38)
as the condition for instability. This can be rewritten as
d lnLd ln T
< 1 (5.39)
for the instability condition; the system is stable other-
wise. This makes it clear that the slope of the cooling
curve in Figure 5.1 controls the thermal stability of the
system. The middle phase of Figure 5.2 lies at a Twhere L is a very slow function of T ; so this phase is
unstable. The outer two phases of Figures 5.2 lie at Twhere L is a steep function of T ; thus these phases are
thermally stable.
References
Most of the formal material here can be found in
• Spitzer (Physics of the Interstellar Medium;
but the thermal stability details come from the original
Field, Goldsmith & Habing paper (1969 ApJ), and I’ve
also pulled more recent arguments (on heating, cooling
mechcanisms) from the current literature.
Key points
• The ISM cooling curve (and what’s inside it);
• Photonionization heating “always” gives ∼ 104 K;
• Stromgren spheres – what they are, what size they
are;
• General ISM thermal balance and why we have the
two cooler phases;
• Why does ISM thermal balance not explain the hot
(coronal) phase?
32 Physics 426 Notes Spring 2016
6 Dynamics of the ISM: energetics & shocks
Last term we worked with the mass and momentum
conservation laws of fluid dynamics, and applied them
to various problems. We now return to fluid dynamics,
and consider the physics of shocks in fluid flow. But
we can’t do that until we look at the third important
conservation law, energy conservation in fluids.
6.1 Fluids: energetics
We must consider two forms of energy: the kinetic en-
ergy density of bulk flows, ρv2/2, and internal energy
density. The latter is the energy contained in random
(thermal) motions of the particles. We will work with
the internal energy per unit mass, e = 1γ−1
pρ . For a sub-
relativistic, monatomic gas, for instance, e = 32kBTm .
The net energy in our volume V is∫
V ρ(e + 12v
2)dV .
The net rate of change of this energy from intrinsic
changes and from flows is
∫
V
∂
∂t
[
ρ
(
e+1
2v2)]
dV
+
∫
V∇ ·[
ρv
(
e+1
2v2)]
dV,
(6.1)
where we have used the divergence theorem to convert
a surface integral to a volume integral as in Chapter
4 of the 425 notes. This net energy change must be
accounted for by (a) work done by an external accel-
eration f , which is often taken to be gravity; (b) work
done by the external pressure; (c) direct energy gains
or losses, most commonly direct heating (by photons
or cosmic rays, say), or radiative losses, as in chapter
5. These three energy-change factors are
∫
Vρf · vdV −
∫
Apn · vdA +
∫
V(Γ− Λ)dV (6.2)
Again we can use the divergence theorem to convert
the pressure work term to a volume integral, and we
can derive one version of the differential energy con-
servation law:
∂
∂t
[
ρ
(
e+1
2v2)]
+∇ ·[
ρv
(
e+1
2v2)]
= ρf · v −∇ · (pv) + Γ− Λ
(6.3)
The forms we derived for the mass and momentum
conservation equations are pretty standard. However,
there does not seem to be one standard form for the en-
ergy conservation equation; rather, one uses the form
that works best in a given application. Therefore, at
the expense of a little algebra, we will look at several
alternate forms of (6.3).
First, with the help of the continuity equation1 we can
separate out the ∂ρ/∂t and ∇ · (ρv) terms in (6.3), we
find
ρ∂
∂t
(
e+1
2v2)
+ ρv · ∇(
e+1
2v2)
= ρf · v −∇ · (pv) + Γ− Λ
(6.4)
which is one alternate form that we will use again. We
can isolate the rate of change of e, from (6.4), by sub-
tracting v·(the momentum conservation equation), giv-
ing
ρ∂e
∂t+ ρv · ∇e = −p∇ · v + Γ− Λ (6.5)
In this expression, we can see that the rate of change of
the internal energy depends explicitly on compression
work (“pdV ” work), and on the net heating and cooling
rates.
Yet another common form of the energy equation is
found by defining the convective, total or Lagrangian
derivative,D
Dt=
∂
∂t+ v · ∇ (6.6)
With this, we can use the continuity equation to write
∇ · v in terms of the density derivatives, and use e =1
γ−1pρ to write
ρ
γ − 1
D
Dt
(
p
ρ
)
− p
ρ
Dρ
Dt= Γ− Λ (6.7)
or, if we collect the p and ρ derivatives separately, we
get
ργD
Dt
(
p
ργ
)
= (γ − 1)(Γ− Λ) (6.8)
which is the last of our alternate forms of the energy
equation.
This last form allows us to consider a couple of impor-
tant limits. The first is the adiabatic limit. If Γ−Λ = 0,
so that there is no net gain or loss of energy to the sys-
tem, (6.8) shows that
p
ργ= constant (6.9)
1You saw this last term, in chapter 4 of the P425 notes. One
form is∂ρ
∂t+∇ · (ρv) = 0
Physics 426 Notes Spring 2016 33
which is the usual adiabatic law (the consequence of
there being no gain or loss of heat from a system).
The second limit is the isothermal limit. A good many
calculations assume T = constant, which simplifies
things enormously. From (6.5), we see that
p∇ · v = Γ− Λ (6.10)
is the condition that must be satisfied if T (or e) is con-
stant.
6.2 Supersonic flow and shock fronts
In P425, we found a characteristic signal speed in a
gas, namely the sound speed, cs. This is a critical find-
ing: because this is the speed at which a perturbation
propagates, “information” about changes in the flow
can only propagate at cs. Figure 6.1 illustrates this, in
a moving flow.
c c ss
reverse waves forward waves
perturbation
Undisturbed flow
Figure 6.1 Illustrating signal propagation in a flow. A per-
turbation “whacks” the pipe at one spot; the information that
this has happened travels upstream and downstream at cs,
relative to the flow. Following Thompson figure 8.6.
Consider, then, gas moving at a speed greater than the
sound speed; if follows that information cannot prop-
agate upstream. This means the gas generally cannot
adjust smoothly to changes in the ambient or bound-
ary conditions, but rather must adjust instantaneously
- creating a discontinuity in the flow. Such a disconti-
nuity is a shock. Examples are bow shocks around su-
personic aircraft, or around the planets (since the solar
wind is supersonic); or standing shocks, such as where
supersonic flow runs into a zero-velocity surface (for
instance, at the end of a radio jet, where it runs into the
ambient plasma).
We treat a shock as an infinitely thin discontinuity in a
flow. The true width of the shock is determined by
collision processes within the fluid, and by assump-
tion these operate on scales much smaller than those
described by the fluid equations. The intent is to de-
rive “jump conditions” – to use the basic conserva-
tion laws to derive relations between the fundamental
variables (ρ, p, T, v) upstream and downstream of the
shock. Let “1” describe upstream, and “2” describe
downstream
v1
n1
p1
1
v2
n2
p2
TT 2
S
upstream = "in" = "out"
Figure 6.2 A close-up look at a shock, S, in a frame in
which the shock is at rest. The incoming fluid is labelled
with subscript “1”; outgoing with subscript “2”. The in-
coming fluid must be supersonic, v1 > cs1. The outgoing
fluid is slower (v2 < v1), denser, n2 > n1), and probably
hotter (T2 > T1, for an adiabatic shock); the incoming ki-
netic energy is converted to heat when the shock decelerates
the flow.
the downsteam flow, as seen in a frame moving with
the shock (as in Figure 6.2). Let the Mach number
be M = v/cs (generally defined for upstream flow).
Consider steady, one-dimensional flow, with no exter-
nal forces, and with no net external heating or cool-
ing. Referring back to the footnote on the prevous
page, the continuity equation for the fluid in steady
state becomes ∇ · (ρv) = 0. If we integrate this over
a small, Gaussian surface enclosing some part of the
shock plane, we get
ρ1v1 = ρ2v2 (6.11)
This is, of course, simply mass conservation: the flux
in (per area) must equal the flux out. The force equa-
tion for the fluid is
∇ · (ρvv) +∇p = 0 (6.12)
(again, go back to your P425 notes, where this form
was presented implicitly, as equation (4.3), but not
elaborated on). Integrating this across the shock face,
we get
ρ1v21 + p1 = ρ2v
22 + p2 (6.13)
These two equations are general. The energy equation
is generally applied in either the adiabatic or isother-
mal limits.
6.2.1 Adiabatic shocks
The form (6.3) of the energy equation, with some alge-
bra (always!) and applied to our conditions here, gives
34 Physics 426 Notes Spring 2016
us
γ
γ − 1p1v1 +
1
2ρ1v
31 =
γ
γ − 1p2v2 +
1
2ρ2v
32
Factoring out ρv from each side, and using (6.13), this
can be written
γ
γ − 1
p1ρ1
+1
2v21 =
γ
γ − 1
p2ρ2
+1
2v22 (6.14)
Now, this can be combined with (6.11) and (6.13), to
express three post-shock quantities, ρ2, v2 and p2, in
terms of their pre-shock counterparts. This solution is:
ρ1ρ2
=γ − 1
γ + 1+
1
M2
2
γ + 1
p2p1
=2γM2 − (γ − 1)
γ + 1v2v1
=γ − 1
γ + 1+
1
M2
2
γ + 1
(6.15)
When M ≫ 1, these equations simplify, to
ρ1ρ2
=γ − 1
γ + 1
p2p1
=2γM2
γ + 1v2v1
=γ − 1
γ + 1
(6.16)
In particular, if γ = 5/3, we find ρ2/ρ1 = v1/v2 = 4(this is often quoted as the strong shock limit). And, in
this limit, the temperature jump is T2/T1 = 5M2/16,
giving kBT2 = 3mv21/32; the upstream kinetic energy
is converted to internal energy in an adiabatic shock.
6.2.2 Isothermal shocks
The energy equation is this case is simple: T2 = T1
by assumption. The possibility of an isothermal shock
depends on the cooling times. We would expect a gen-
eral shock to have a structure as in Figure 6.3. The
gas passing through the shock is initially heated, by
adiabatic compression, and suffers a moderate density
jump. This hotter gas (assuming an optically thin sit-
uation), can then cool by radiation, and while cool-
ing the gas travels a distance ∼ v2tcool. This “cool-
ing distance” thus measures the effective width of the
transition to an isothermal shock – assuming that some
heating/cooling balance, as we described for the ISM,
maintains the upstream and far-downstream tempera-
ture at T1.
T T
n
n 1 T 1
n
v1
v 2
n2 T2
(transition region)
Figure 6.3 Schematic diagram of a radiating shock. In the
upper diagram, the fluid is coming in from the left. At x1 it
enters the nonradiative shock. This is followed, to the right,
by the transition region, where the temperature drops as the
gas cools by radiation. The changes of density and tempera-
ture are shown schematically in the lower figure. Following
Spitzer figure 10.1.
Combining T2 = T1 with (6.11) and (6.13), to express
isothermal jump conditions,
ρ2ρ1
= M2
v2v1
=1
M2
p2p1
= M2
(6.17)
Thus, the compression factor, and deceleration factor,
can be much higher for an isothermal shock than for an
adiabatic one.
6.2.3 Magnetized shocks
The previous analysis ignored the magnetic field. This
is probably too naive, as we know the ISM is magne-
tized. We can understand the effect of a shock on the
field by considering flux freezing. Refer to the left part
of Figure 6.4. In this case, the field is parallel to the
shock face. In this geometry, the field is tied to the gas
by flux freezing; thus the field behind the shock will
be increased, in the same amount as the gas is com-
pressed. By comparison, consider the right part of the
figure, in which the shock propagates along the field
lines. In this case, the density jump will not affect the
magnetic field (why? can you see how this is consistent
with flux freezing?).
6.2.4 Oblique shocks
Finally, what if the shock face is not perpendicular to
the flow? The answer is qualitatively simple, and can
be readily understood by thinking about an oblique in-
Physics 426 Notes Spring 2016 35
S S
B B B B
v v ss
1 2
quasi−perpendicular shock quasi−parallel shock
Figure 6.4 Schematic picture of magnetized shock, in
two important limits The shock speed relative to the medium
is vs. Left, quasi-perpendicular shock (note vs ⊥ B): flux
freezing increases the post-shock field, by a factorB2/B1 =ρ2/ρ1. Right, quasi-parallel shock; the gas flows along the
field lines without perturbing the field.
coming velocity (w in the Figure 6.5, which illustrates
the geometry) in terms of its components parallel and
perpendicular to the shock face. The component per-
pendicular to the shock face is decelerated, just as in
the normal-shock results above. The component par-
allel to the shock face, however, is not affected (query
to the reader: why not?). Thus, the net velocity bends
toward the shock face.
v1
1
v 2
u2
w1
w2
u
β
β2
w1
w2
β
δ
Figure 6.5 Oblique shocks. The velocity bends towards
the shock face, as shown in the right figure; this can be un-
derstood by considering the effect of the shock on the veloc-
ity components, as in the left figure.
Where do we expect to deal with oblique shocks? In
just about any 2D situation ... a good example is the
terrestrial bow shock, where the supersonic solar wind
encounters the earth. The flow coming in bends to-
ward the shock normal, and thus is deflected around
the earth. The jump conditions (extensions of 6.15 or
6.17) can be quite complicated algebraically; we won’t
deal with them in this course.
Key points
• Energy conservation: adiabatic and isothermal limits
• Shock fronts: jump conditions across the shock
• Shock fronts: adiabatic and isothermal limits
• Effects of B field or oblique angle: qualitative
36 Physics 426 Notes Spring 2016
7 Stellar Winds & Supernovae Remnants
Now let’s look at some supersonic flow situations
which involve shocks. You remember that we worked
with smooth transonic flows last term – for instance
the solar wind. These are rare; it’s very easy to form
shocks when supersonic flows are decelerated or bent
by their surroundings. In this chapter we’ll work with
2-1/2 examples: stellar wind shocks, and two types of
supernova remnants.
7.1 Stellar winds and the surrounding ISM
We worked with smooth, transonic stellar wind flow
last term ... and found a solution in which the outer
regions of the wind flow are supersonic. But this can’t
carry on forever. At some point the wind flow must run
into the ambient ISM. We expect the deceration to lead
to an outer shock in the wind; what are the details? By
way of review, I’ll repeat the basics of the inner-wind
solution here.
7.1.1 The basic solution
• Mass conservation in a steady, spherical flow is
ρvr2 = constant; or,
1
ρ
dρ
dr+
1
v
dv
dr+
2
r= 0 (7.1)
while the momentum equation becomes in this case
(noting that gravity from the central star is important),
ρvdv
dr+
dp
dr= −ρ
GM
r2(7.2)
Writing dp/dr = c2sdρ/dr, these two equations com-
bine to give the basic wind equation,
(
v − c2sv
)
dv
dr=
2c2sr
− GM
r2(7.3)
This does not have analytic solutions over the whole
range of r. However, we can learn quite a bit about the
nature of the solutions simply by inspection of (7.3), as
follows.
• First, the left hand side contains a zero, at v2 = c2s .
If we want to consider well-behaved flows, that is to
say those in which the derivative dv/dr does not blow
up, then the right hand side of (7.3) must go to zero at
the same point. This defines the condition that must be
met at the sonic point:
v2 = c2s at r = rs =GM
2c2s(7.4)
Whether or not a particular flow satisfies this condition
depends on the starting conditions, such as with what
velocity and temperature it left the stellar surface, and
also what the boundary conditions at large distances
are. If it does not start in such a way to satisfy this
condition, it either stays subsonic (corresponding to fi-
nite pressure at infinity), or cannot establish a steady
flow.
• Further, the solution beyond the sonic point depends
on the temperature structure of the wind. The only so-
lutions with dv/dr > 0 for r > rs are those for which
c2s(r) drops off more slowly than 1/r; it is only these
for which the right-hand side stays positive. In the case
of an isothermal wind, with c2s = constant, (7.3) can be
solved in the limit r ≫ rs:
v2(r) ≃ 4c2s ln r + constant (7.5)
Thus, the wind will be supersonic, by a factor of a few,
as r → ∞. The question of how the solar wind man-
ages to stay nearly isothermal is not solved; it is prob-
ably due to energy transport by some sort of waves
(MHD or plasma waves, for instance) which are gen-
erated in the photosphere and damped somewhere far
out in the wind.
7.1.2 The outer shock
The pressure in this supersonic wind is dropping with
radius (since ρ ∝ 1/vr2, with v being only slowly
varying; thus p ∝ ρT ∝ 1/r2, approximately, in
an isothermal wind). Therefore, when the wind pres-
sure is close to the ISM pressure, the wind must slow
down. At this outer boundary, we expect some sort of
shock transition, since the wind is supersonic. Past this
shock, the hot, shocked wind-gas will expand into the
ISM (at about its own sound speed, to start); as long
as this expansion is supersonic relative to the ISM, the
expanding hot gas will drive a “snowplowed” shell of
ISM, and a second shock, out into the ISM.
A cartoon of this region, at some point in time, would
be that in Figure 7.1. Let region “a” be the wind; S1
be the inner shock; region “b” be the wind-gas which
has been through the shock; C be the contact sur-
face between the wind and the ISM; region “c” be the
shocked ISM; and S2 be the outer shock (moving into
the ISM). We expect S1 to be an adiabatic shock (since
the wind is probably hot and low density, and thus will
have a long cooling time); region “b” will contain hot,
shocked wind, with Tb ∼ 316
mv2windkB
∼ several×107 K
Physics 426 Notes Spring 2016 37
(noting that m = 12mp is the mean mass per particle if
region “b” is fully ionized). The outer shock will prob-
ably be isothermal, since the ISM is denser and cooler
than the wind. Thus, the shocked ISM will be in a
thin shell, containing all of the original ISM that lay
between S2 and the star.
S
C
S 2
1
R(t)
a
bc
d
Figure 7.1 Cartoon of the structure of a stellar wind, and
its interaction with the ISM. Left: the shock structure within
the wind. Right: The outer shell of dense, snowplowed ISM.
From Dyson & Williams figures 7.3 and 7.4.
To start, consider the position of this shell, R(t), as a
function of time. We start with a force equation. The
mass is the ISM shell is the mass that was originally
within R(t): 4π3 ρoR
3(t). The force acting on the shell
is just the pressure behind it, which drives it outward:
d
dt
(
4π
3R3ρo
dR
dt
)
= 4πR2pb (7.6)
if pb is the pressure, in the outer part of region “b”,
which acts on the shell. Next, we need an equation
for pb(t). The energy within region “b” – which we
will take to be nearly all of the volume within R(t)– is ≃ 2πpbR
3(t); the rate of change of this energy
is given by the difference between the input (from the
wind) and the “pdV” work done by the expansion:
d
dt
(
2πR3pb)
= Ewind − pbd
dt
(
4π
3R3
)
(7.7)
if Ewind is the energy input rate from the wind, as-
sumed constant. These two equations can be com-
bined, for instance by eliminating pb, to get
aR4d3R
dt3+ bR3dR
dt
d2R
dt2+ cR2
(
dR
dt
)3
=3
2π
Ewind
ρo(7.8)
where a, b, c are numerical constants. Noting that each
term on the left hand side has the same dependence on
R and t (∼ R5t−3), we can try a power law solution,
R(t) = Atα. A small bit of algebra tells us that α =
3/5 is the allowed solution, and we can also find an
expression for A. Thus, the solution is
R(t) = 0.76
(
Ewind
ρo
)1/5
t3/5 (7.9)
and
v(t) =dR
dt= 0.46
(
Ewind
ρo
)1/5
t−2/5 (7.10)
Thus, the shell decelerates with time, as it ought to if
it is picking up more and more ISM. One can then use
these, and (7.6), to work out the pressure acting on the
outer shell:
pb(t) =7
25A2ρot
−4/5 (7.11)
so that the outer pressure drops with time. Finally, one
can also work out the kinetic energy of the shell, which
must give the energy input to the general ISM from the
wind. This turns out to be
2π
3R3ρo
(
dR
dt
)2
≃ 0.2Ewindt (7.12)
so that about 20% of the wind energy goes to the ISM.
(The rest goes to heating the bubble, and to “pdV”
work).
7.1.3 What about the inner shock?
The location of the inner shock, S1, is determined by a
combination of the jump conditions, applied at S1, and
the pressure at the outside of the region, pb, as follows.
• The jump conditions, at S1, are ρsb = 4ρsa, if “sb”
and “sa” subscripts refer to postshock (region “b”) and
preshock (region “a”), respectively. Also, vsb = vsa/4– here we assume M ≫ 1, the strong shock limit,
and also an adiabatic shock. At the shock, momentum
conservation1 tells us that ρv2 + p is conserved; thus,
psb =3
4ρsav
2sa + psa ≃ 3
4ρsav
2sa
where we have used the fact that vsa ≫ cs in the wind
region.
• In region “b”, where we can ignore gravity (com-
pared to the internal energy, 32p), the momentum equa-
tion is
ρvdv
dr+
dp
dr≃ 0
1Check back to chapter 4 from last term, P425; equation 4.4.
38 Physics 426 Notes Spring 2016
and, since v ≪ cs in most of region “b”, ρ ≃ con-
stant, and we have 12ρv
2 + p ≃ constant. Now –
we know that v drops from vsb = vsa/4, at S1, to
vs2 ∝ t−2/5 ≪ cs,sb at S2 (from the shock conditions).
Thus, we connect the conditions just past S1 to the con-
ditions at S2 (remembering that we called the pressure
in region “b” at S2, pb; ugly notation, I agree!),
pb ≃ psb +1
2ρsbv
2sb =
7
8ρsav
2sa
Thus, pb, at S2, is set by the dynamic pressure at S1.
If pb is also set by external conditions – say the ISM
pressure – then the location of S1 must be where ρsav2sa
satisfies the above condition.
• But we can find this location. We note, again, that the
dynamic pressure in the wind region, ρv2 ∝ Mv/r2
drops approximately as 1/r2 (since v is slowly vary-
ing). Thus, the behavior of the dynamic pressure
within the wind region is fixed by the basic wind so-
lution. Thus, the shock S1 must form where ρv2, as
determined by the wind solution, matches ∼ 87pb.
7.2 Supernova remnants
First, set the stage: a star explodes. You probably recall
the basic picture: stars meet a violent death about 4-5
times per century in a galaxy the size of ours. There
are two possible types of supernovae. Type I super-
nova (more correctly Type Ia) arise from the explosion
of a white dwarf in a close binary system, presumably
initiated by sudden mass transfer from the companion
which leads to a thermonuclear reaction in the dwarf
star.2 Type II supernovae come from evolved, mas-
sive stars in which nuclear burning has run out of fuel.
The stellar core collapses inwards and bounces, pro-
ducing the explosion.
For the purposes of these notes, both types of SNe re-
sult in very similar remnants. The dynamics of the
ejected material will be sightly different at very early
times, due to the differing local environments; but af-
ter a short time, both can be treated similarly. That
is the approach we will take here. Think of an in-
stantaneous release of energy (E ∼ 1051 ergs) from
a point source, in ambient gas of density ρo. The en-
ergy released will heat the gas near the explosion to
2Current work splits this hair. The distinction between Type
I and Type II SNe was originally based on the presence, or ab-
sence, of strong hydrogen lines in the explosion spectra. Origi-
nally, all stars without H lines were thought to be explosions of
white dwarfs; now I gather Type Ib and Ic are identified with core
collapse.
very high temperature and pressure, driving an expan-
sion. This expansion will be very supersonic, setting
up a spherical shock wave moving into the surround-
ings and sweeping up gas as it goes.
This picture is similar to our previous model of an ex-
panding stellar wind bubble; but different in some re-
spects. First, obviously, is the δ-function nature of
the explosion. We expect that to change details, for
instance the power-law evolution of the radius of the
shock. In addition, we have to consider the radiative
cooling rate at the outer shell. With stellar winds, we
argued that the outer shock is radiative (isothermal);
this is justified (after the fact) by the high densities and
slow expansion speeds of the wind system. For SNR,
however, the situation is different. They can occur in
lower density regions (for instance an HII region?), and
have much higher explosion speeds (thus much hot-
ter post-shock temperatures). We must consider two
phases, then: (a) an early energy-conserving phase,
during which radiative losses are unimportant, and (b)
a later snowplow phase, in which the shell becomes
dense and cool, and the remnant evolves by momen-
tum conservation.
7.2.1 Early: energy conserving (Sedov) phase.
The remnant in this stage can be thought of as a
large hot bubble, filled with ambient gas that has been
through the outer, strong shock. Again, let the radius of
the outer shock be R(t). In this case, both the internal
and kinetic energies per mass are given by
e =1
2v2 =
9
32R2 (7.13)
This is a direct post-shock result, and holds in a fixed
reference frame – one in which the shock is advancing
into a medium at rest. We will also assume that gra-
dients within the bubble are small so that (7.13) holds
throughout. ( This latter is an OK, but not wonderful,
assumption – cf. Figure 7.2.) The total energy is then
Etot =4π
3R3ρo
(
e+1
2v2)
=3π
4ρoR
3R2 (7.14)
But now, we require Etot = ESN , the input energy
from the SN. We thus have an equation of motion for
the shock:
R3R2 =4
3π
ESN
ρo(7.15)
This solves to
R(t) =
(
25
3π
)1/5 (ESN
ρo
)1/5
t2/5 (7.16)
Physics 426 Notes Spring 2016 39
and
V (t) = R(t) =2
5
(
25
3π
)1/5(ESN
ρo
)1/5
t−3/5
(7.17)
Compare these to (7.9, 7.10) for a stellar wind: note
the different functional dependence on time.
To go further, the basic fluid equations can be used to
determine the structure of the interior of the hot bubble.
It is the well-known Sedov-Taylor solution. I do not
reproduce it here, but show the results (which must be
done numerically) in Figure 7.2.
Figure 7.2 Solution for the interior structure of the hot
bubble in the energy-conserving phase of the SNR. The ra-
dius is scaled to rsh = R(t) in our notation; this solution
preserves its functional shape as the remnant expands. From
Shu Fig. 17.3.
7.2.2 Late: momentum conserving (snowplow)
phase.
At some point the outer shell will cool; as its pressure
drops it will be compressed by the hot, expanding inner
gas. This is reminiscent of the outer shell in the wind
case, but with differences. Here, the dense shell con-
tains only a small fraction of the ISM which was ini-
tially inside R(t). In addition, the interior hot bubble
is not receiving ongoing energy input, so the pressure
on the outwards shell is dropping quickly with time.
The basic analysis of this late-time remnant, then, is
based on the simple assumption of momentum conser-
vation. Say the remnant enters this phase at some time
to, when it has radius Ro and velocity Ro. Conserva-
tion of momentum has,
4π
3R3ρoR = constant =
4π
3R3
oρoRo (7.18)
and this integrates to
R(t) = Ro
[
1 + 4Ro
Ro(t− to)
]1/4
∝ t1/4 (7.19)
and
V (t) = R(t) = Ro
[
1 + 4Ro
Ro(t− to)
]−3/4
∝ t−3/4
(7.20)
where the last proportionality statements are valid for
t ≫ Ro/Ro. Thus, comparing this to (7.16,7.17)
shows that in the momentum-conserving phase, the
remnant expands more slowly . . . as one would ex-
pect, right?
Dyson & Williams present some numbers. The Sedov
phase has R(t) ≃ 3.6 × 10−4t2/5 pc, and V (t) ≃4.4 × 109t−3/5 km/s (if t is in seconds). Strong cool-
ing typically takes over for Ro ∼ 250 km/s, giving
Ro ∼ 24 pc and an age ∼ 30 × 104 years. At that
point, about 1400M⊙ of ISM has been swept up –
much larger than the mass initially ejected (something
like 4M⊙). Thus, we are looking almost entirely at the
dynamics of the ISM which was dramatically heated in
the star’s explosion.
7.3 Plerions, a.k.a. pulsar wind nebulae
The preceding section discussed the “classical” picture
of supernova remnants, in which energy is injected at
one instant into the ambient ISM. Many galactic SNR
are well described by this model. But not all: in some
cases the exploding star leaves behind an active pul-
sar as its remnant. We now know that many (most?
all?) pulsars drive a relativistic wind out from the star.
The wind acts as a source of mass and energy, fill-
ing the interior of the SNR with hot (or relativistic)
plasma. These “filled” SNR used to be called pleri-
ons (mostly in the SNR community); these days they
are being called pulsar wind nebulae (in the ISM/X-
ray community). As far as I know the two terms refer
to the same type of object.
We have direct evidence of pulsar winds. For older
pulsars (those not currently within SNR), we see struc-
tures in the nearby ISM which are clearly bow shocks
associated with the star’s high-speed motion through
the ISM. From the standoff distance of the bow shock
we can estimate the wind energy, and compare it to
standard models of the pulsar. For young pulsars (those
still within their SNR), recent CHANDRA images di-
rectly reveal the outflow from the pulsars (at this point
there is a good handful of such images; the Crab and
Vela pulsars are the most famous). These outflow are
complex: they show jets, which presumably come out
along the star’s rotation axis (this is the only symme-
40 Physics 426 Notes Spring 2016
try axis in the system), and equatorial winds, which
probably arise from the combined effects of the star’s
strong magnetic field and its rapid rotation.
When a relativistic wind hits the local ISM (which may
be the ejecta from the SN), we expect strong shocks
to form. Such shocks may be effective at accelerating
particles in the local plasma to relativistic energies (we
will discuss this in detail later in the course). Thus,
the material which has been through the shock may
contain a large fraction of relativistic particles – which
could, for instance, maintain the nonthermal emission
from the surrounding nebula (Crab or Vela). The dyan-
mics of such a system – assuming it’s close to spheri-
cally symmetric – are of course very similar to the dy-
namics of a stellar wind hitting the ISM.
Key points
• Solar-wind solutions (from last term);
• The outer wind shock and getting its location from
dimensional analysis;
• Supernova remnants, Sedov and snowplow phases;
• Plerions, what they are, how they work qualitatively.
Physics 426 Notes Spring 2016 41
8 Relativistic particles in astrophysics
Based on our knowledge of cosmic rays and, less di-
rectly, of the relativistic electrons which radiate in
synchrotron sources, our general picture is of parti-
cles which are highly relativistic – that is, with γ =E/mc2 ≫ 1 – but which are also tied to the back-
ground, thermal plasma in which they find themselves.
The distribution function of these particles is found to
be a power law, rather than a Maxwellian; thus, these
particles cannot have had time to thermalize (in the
two-body sense). In this section, we will consider the
dynamics of these particles and how they are tied to the
background plasma). Later on we’ll address how they
may be accelerated.
8.1 Recap: basics for relativistic particles
First, basic special relativity. The total total energy of
a relativistic particle is given by E2 = p2c2 + m2c4;
we also have the definition E = γmc2, where γ2 =1/(1 − β2), and β = v/c. In the limit E ≫ mc2, we
also have E ≃ pc (which is exactly true for a photon,
of course; as the particle gets more relativistic, its rest
mass becomes less and less important). Note that the
Lorentz factor γ is often used to represent the energy;
the mc2 factor is implicitly carried along if you need
“real” units.
Power-law distribution functions (motivated by cosmic
rays) are often used:
f(E) = foE−s , E1 ≤ E ≤ E2 (8.1)
or
n(γ) = noγ−s , γ1 ≤ γ ≤ γ2 (8.2)
The exponent s depends on the system; the scaling
constant fo or no connects to the total number (or num-
ber density) of particles. That is, the total number of
particles will be N =∫
f(E)dE =∫
n(γ)dγ. It fol-
lows, then, that f(E) and n(γ) have different units –
watch out, this difference can bite you.1
1To be specific: the two DF’s are equivalent if
f(E)dE = n(γ)dγ ; f(E) = n(γ)dγ/dE
(so that we have the same number of particles “at E = γmc2”).
For the specific power law case, above, this gives
foE−sdE = noγ
−sdγ ; fo = no(mc2)s−1.
Remaining question to the reader: how is no related to the total
number of particles N?
8.2 Quick overview of the observations
We have seen that astrophysical plasmas contain two
species. One species is the thermal interstellar gas
which we have been considering. In this context, its
important properties are that it seems to be well de-
scribed by a thermal equilibrium distribution function
(a Maxwell Boltzmann velocity distribution), and that
the energy per particle is subrelativistic. (Recall tem-
peratures range from O(10)K to O(106)K).
In addition, many astrophysical plasmas – including
the galactic ISM – contain a significant population of
highly relativistic particles which are not in a thermal
distribution. We saw last term that we have direct and
indirect evidence of these particles; I’ll review the ar-
guments here.
Baryons. Here we have direct evidence – these are
the cosmic rays. They are mostly protons, but there is
a heavy element component, with approximately solar
abundances (so they come from processed material).
They are very isotropic in arrival direction, probably
at all energies (although arrival directions for the very
highest energy particles remain uncertain).
• The baryon energy distribution is a power law,
N(E) ∝ E−s, with a break at E ∼ 1015 eV (the
“knee”), and another at E ∼ 1019 eV (the “ankle”).
The exponent s ∼ 2.7 below the ankle, and higher
above. Comparison of the gyroradius to the scale of the
galaxy suggests that the highest energy CR, above the
ankle, are extragalactic, while the lower energy ones
are galactic in origin.
Leptons. Here we have some direct evidence – the
lepton component in the cosmic ray spectrum can be
separated from the baryon component. The cosmic ray
lepton distribution falls much more steeply than the
baryon distribution, above energies ∼ 1 GeV, so that
its total contribution to the CR energy density is only
∼ 1% that of the baryons.
In addition, there is also abundant indirect evidence
fore highly relativistic electrons2 throughout the uni-
verse. Synchrotron radiation – which we will study in
detail in the next chapter – is common in many dif-
ferent settings. Because synchrotron radiation comes
from highly relativistic particles (with γ = E/mc2 ≫1), we know immediately that synchrotron sources
have a relativistic lepton component. Examples of this
2Well, really leptons; we’ll discuss later whether we can distin-
guish electrons from positrons by their radiation signatures.
42 Physics 426 Notes Spring 2016
are the galactic disk; supernova remnants, both stan-
dard and “filled”; radio jets from compact stars in X-
ray binaries; X- and γ-ray emission from pulsars; ra-
dio jets from active nuclei, and the radio lobes they
create; diffuse synchrotron emission from the plasma
in clusters of galaxies (including a few synchrotron-
bright shocks created when two clusters collide); and
quasars of course.
Looking ahead to the next chapter, we can note some
important characteristics of synchrotron radiation. It
requires magnetic fields and highly relativistic elec-
trons. The spectrum from a single particle, with energy
E = γmc2, peaks at a photon frequency
νsy ∼ 3
4πγ2
eB
mc(8.3)
Thus, for a uniform B field, one particle energy γ maps
directly to one (observed) photon frequency, ν. This
single, radiating particle has an energy loss rate
dE
dt≃ 4
3cσTγ
2B2
8π(8.4)
where σT = 6.65 × 10−25cm2 is the Thompson cross
section. Synchrotron radiation commonly has a power
law spectrum, which tells us (assuming the B field is
simple) that the underlying electron distribution is also
power law (just as it is in galactic cosmic rays). Typical
values for the photon spectrum are j(ν) ∝ ν−α, with
α ∼ 0.5 − 1.0; the associated electron distribution is
n(γ) ∝ γ−s, with s ∼ 2.0 − 3.0.
From(8.4), we see that higher energy electrons lose
energy faster that the lower energy ones, because
dE/dt ∝ E2; from this one can show that an initial
electron power law spectrum will develop a break at
the energy where the particle’s radiative lifetime equals
the age of the plasma. As the plasma gets older, this
break will move to lower energies (and thus to lower
photon frequencies).
8.3 Cosmic rays in the galactic setting
One of the big questions is, how are cosmic rays accel-
erated to such high energies, and how are they main-
tained there (why don’t they eventually thermalize with
the ISM)? Before we go there, let’s recap a few points
about cosmic rays from last fall (P425 notes).
Sources are thought to be two-fold. Supernova rem-
nants have long been thought to be the main source of
CR; I personally suspect that pulsars are probably also
an important source. Still another possibility is that the
highest energy CR may have an extragalactic origin.
Propagation and Trapping. Once generated, CR do
not just fly freely through space. Becuase they are
charged, they are connected to the ISM by their gyro-
motion, and by scattering on turbulent Alfven waves
in the ISM. Thus the CR distribution we observe at
earth may well have been seriously changed, relative
to their “birth” distribution, by propagation and scat-
tering through the ISM on their way to us. From nu-
clear abundances we learn that galactic CR typically
spend most of their life in the galaxy, not in the disk,
but rather in the more extended halo; and that its life-
time to escape from that halo ∼ 20 Myr.
Losses. The leptons, being of smaller mass, are sus-
ceptible to radiative losses (synchrotron in the galactic
magnetic field, inverse Compton scattering on what-
ever radiation is around) as well as Coulomb losses
(scattering on the plasma component of the ISM). This
also modifies the electron energy distribution, com-
pared to the source, and of course reduces the net en-
ergy in the electron component of the CR.
8.4 Particle acceleration, overview
Next, we consider the question of particle accelera-
tion: what is the origin of cosmic rays (which we ob-
serve directly at earth), and of the relativistic electrons
in such sources as supernova remnants and radio jets
(which we observe indirectly via their synchrotron ra-
diation)? How are these charged particles accelerated
to relativistic energies, and how are they maintained in
a non-Maxwellian distribution? The short answer is,
this must be done by electric fields, E 6= 0. Magnetic
fields do no work on the particles, and gravity is a con-
servative force (so that any energy a particle gains by
going into a gravitational potential well must be lost
again when it leaves the well). However, is it not easy
to maintain large-scale electric fields in space, where
the abundant free charges in astrophysical plasmas will
want to short out any static field. Two types of particle
acceleration models exist — which I tend to call “first
stage” and “second stage” mechanisms.
8.5 Particle acceleration, first stage mechanisms
One possibility is that an ordered, large-scale E field is
maintained in the region of some massive object, like a
star (thinking of solar flares) or a pulsar or an accretion
disk, where dynamic processes can maintain a strong
Physics 426 Notes Spring 2016 43
dynamo; these are called first stage mechanisms. In
this situation there is no minimum energy threshold for
particle acceleration; any charged particle dropped in
a region with E 6= 0 will be energized. There may,
however, be an upper limit to the energy that can be
reached – due to the physical size of the system (and
the consequent limit on the overall potential drop,∫
E ·ds). There are a couple basic types of models here:
reconnection sites and unipolar dynamos.
8.5.1 Magnetic reconnection
You saw this last term; I’ll just store a recap here. Con-
sider a region in a plasma in which the direction of the
magnetic field reverses over a small spatial scale – for
instance, the neutral sheet in the earth’s magnetotail,
or the base of a flux tube footed in the solar photo-
sphere. The magnetic field in this region will find a
lower-energy state by “reconnecting”, that is by chang-
ing the magnetic field configuration. Such a region can
be the site of rapid conversion of stored magnetic en-
ergy to kinetic and internal energy of the plasma, Re-
connection is believed to be important in solar flares,
and has been observed in the earth’s magnetotail.
V
V
B
B
B
BB
B
V V
V
V
Figure 8.1 The geometry of magnetic reconnection. Left,
a simple system before reconnection; the field lines have op-
posing directions and meet in a central current sheet (called
a neutral sheet), shaded in this figure. Right, with reconnec-
tion ongoing; some field lines have changed their topology,
and plasma is driven out of the neutral sheet across the re-
connected field lines. Think: which way do the current and
E field point in the neutral sheet?
The reason reconnection is important for particle ac-
celeration, is that the neutral sheet (where the magnetic
field goes through zero) contains an ordered electric
field, of strength
E = − c
4πσ∇×B− 1
cv ×B (8.5)
(This is just Ohm’s law, combined with Maxwell’s
equations; check the section earlier on flux freezing).
Here, σ is the electrical conductivity (not a cross sec-
tion). In the cartoons above, the E field will be in the
plane of the neutral sheet and perpendicular to the pa-
per. The maximum particle energy that can be reached,
then, is set by the potential drop across the neutral
sheet, ∆V =∫
E · dl.
8.5.2 Unipolar dynamos
This is a jargon-word for massive, rotating, compact
objects – pulsars and accretion disks – which can sup-
port strong fields. (Refer back to Fig 9.3 of your P425
notes.) If a magnetic field is tied to the rotating matter
or plasma (say by a high conductivity, so that the field
is approximately flux-frozen), the rotation will cause a
field E = 1cvrot ×B to be seen by a local observer.
The numbers that are in principle reached by rotat-
ing compact objects are impressive. For pulsars, flux-
freezing considerations estimate B ≃ 1012 G if the
field was originally the stellar field. The implied elec-
tric field can accelerate a particle to energies of 1016
eV if it operates over a distance ≈ Rns ≈ 10 km.
For an accretion disk around a black hole, the relevant
length scale is rg = GMbh/c2, the magnetic field can
be 103 G, and the inner parts of the accretion disk ro-
tate at nearly lightspeed. In this case the potential drop
can reach 1019 V.
On the other hand, it is not obvious that these high
fields can be maintained – in fact it’s very unlikely.
The picture we have presented of a rotating, magne-
tized object is a vacuum argument. But the atmo-
spheres of such objects are likely to have some amount
of free charge, which may well short out a large part
of these maximum predicted potential drops. For one
example, think about a relativistic particle being ac-
celerated along the magnetic field line in a pulsar by
these strong E fields. As it gets accelerated, it radiates
– and the γ-ray photons it radiates can pair produce
in the pulsar’s strong magnetic field. The newly cre-
ated electron-positron pairs shield most of the rotation-
induced E field ... leading to a net particle energy only
∼ 10−6 or so of the maximum predicted by the simple
order-of-magnitude estimates.
These two families of models, reconnection and unipo-
lar dynamos, represent most of the models of particle
acceleration by large-scale electric fields. What are
the important characteristics of these models, as far
as their relevance to particle acceleration? First, all
charges which see E are accelerated. Second, the max-
imum energy that can be reached is given by the po-
tential drop; but real-world effects, mostly related to
44 Physics 426 Notes Spring 2016
local plasmas (shorting out part of the potential drop,
or providing turbulence which can stop the speeding
particles before they reach E ∼ e∆V ). Third, due
to the effects just listed, much of the available energy
goes to heating the plasma rather than accelerating a
small fraction of the charged particles to high energies.
The final values of the output particle energies depend
on these details, of course. Observations of solar and
magnetospheric reconnection suggest that some frac-
tion of the charged particles are, indeed, accelerated to
well above thermal energies; but not to relativistic en-
ergies. (The output energy may well be larger in pul-
sars and galactic-nucleus accretion disks; but we can
only work from inference there). Thus, these mecha-
nisms are often thought of as a “first stage” in the ac-
celeration process; stochastic methods may take these
“seed” particles and boost them to much higher ener-
gies.
8.6 Particle acceleration, second stage
mechanisms
The other class of acceleration mechanisms invokes
stochastic electric fields, such as are found in plasma
turbulence. If the particles couple efficiently to the tur-
bulent E fields, they can gain energy from the turbu-
lence. There are again a couple of versions of this: true
turbulence (disordered plasma motions, for instance
somewhere in the dynamic ISM), or shock acceleration
(using shock physics to drive the turbulence).
The fundamental idea of stochastic particle acceration
was invented by Fermi in 1949; his physical picture
was rather naive, but described the basic stochastic
mechanism quite clearly. More recent work has im-
proved the physical description of the scattering mech-
anism, while retaining the basic idea.
We have already seen the most likely scattering mech-
anism: resonant interaction with Alfven waves. In this
situation there is a minimum particle energy which can
interact with the waves, as we saw last term. There-
fore, these models can be thought of as second stage
mechanisms, in that they take “seed” particles created
by first-stage mechanisms and accelerate them to very
high energies. The maximum energy which can be
reached here is also finite, and again tied to the size
of the system (which determines the maximum Alfven
wavelength which can exist, and also determines the
rate at which particles can leak out of the accelera-
tion region). But these upper limits can be higher,
in some astrophysical settings, than those provided by
first-stage mechanisms.
8.6.1 Fermi acceleration
This is the original picture. Think back to last term,
when we talked about “magnetic mirrors”. That is;
for a particle moving in a region of spatially changing
magnetic field, the magnetic moment, µ = p2⊥/2γmBis conserved, as long as the time during which the par-
ticle sees the field to change is long compared to the
particle’s gyroperiod. But since p2 = p2⊥ + p2‖ is fixed
in the absence of external forces, p2⊥ ≤ p2. Thus, there
is a maximum value of B which is allowed; the particle
is kept out of high-field regions. This effect is called
magnetic mirroring.
Now, to apply this to cosmic rays, envision a field of
randomly moving mirrors (we can think of them as in-
terstellar clouds, or as clumps of high magnetic field)
on which the relativistic particles or cosmic rays scat-
ter elastically. We want the effect of many collisions
with these mirrors, on a particle distribution. Let the
mirrors have random velocity vm, and average spac-
ing L. If a particle is moving faster than the mirrors,
it can undergo either overtaking or head-on collisions.
In a single collision, the particle (mass m, velocity v)
suffers velocity and kinetic energy changes,
head− on :
∆v ≃ 2vm; ∆E ≃ 2mvm(v + vm)
overtaking :
∆v ≃ −2vm; ∆E ≃ 2mvm(v − vm)
(8.6)
But, the rate at which the particle suffers headon colli-
sions is ≃ (v+ vm)/L; its rate of overtaking collisions
is ≃ (v − vm)/L. Thus, the net rate of change of the
particle’s energy is
dE
dt≃ (v + vm)
L2mvm(v + vm)
− (v − vm)
L2mvm(v − vm)
(8.7)
and this collects to the final form,
dE
dt≃ 16
E
tcoll
(vmv
)2(8.8)
where we have written tcoll = L/v. While I wouldn’t
take the factor 16 too seriously, the fundamental form
is: dE/dt ∝ E/tcoll, which is the standard result for
Fermi acceleration.
Physics 426 Notes Spring 2016 45
A couple of features are worth noting. One, this is a
test particle approach; the energy loss of the mirrors is
not taken into account. Two, the fractional energy gain
per collision is small, since vm ≪ v ≃ c; thus, the
acceleration time tacc ∼ E/(dE/dt) ∼ (v/vm)2tcoll.
What particle spectrum is predicted by this simple
model? One way to find this is as follows. From the
kinematics above, we found that the average energy
gain per collision is ∆E ∼ E(vm/c)2 if the parti-
cles are relativistic. Thus, after p collisions, a parti-
cle which started at Eo will have energy Ep after pbounces:
Ep ≃ Eo
(
1 +v2mc2
)p
(8.9)
which can be written,
lnEp
Eo= p ln
(
1 +v2mc2
)
≃ pv2mc2
. (8.10)
Now, let the particles have some chance, η, of escaping
from the acceleration region (something must end the
acceleration process, after all!). If the escape time ∼τ , then η ∼ tcoll/τ . But, the number of particles at
Ep, Epf(Ep), is just the starting number, No, times
the probability of the particle staying in the system for
p bounces:
Epf(Ep) = NoP (p) = No(1− η)p (8.11)
which we can write as
ln[Epf(Ep)] = constant + p ln(1− η)
≃ const− pη(8.12)
Combining this with (8.10), eliminating p and drop-
ping the p subscript, we find the predicted spectrum:
f(E) ∝ E−(1+ηc2/v2m) (8.13)
Thus, this model – Fermi acceleration plus a constant
probability of escape from the system – results in a
power law spectrum, as is observed in cosmic rays and
elsewhere.. However, as this model stands, the expo-
nent of the power paw, s = 1 + ηc2/v2m, is a sensitive
function of the mirror parameters and the escape time.
This is not very appealing, since the observed values of
s fall in the range 2 <∼s <∼3 almost everywhere (cosmic
rays, supernova remnants, radio galaxies, etc.).
8.6.2 Plasma turbulence: Alfven waves
Modern work has replaced Fermi’s picture of moving
magnetic mirrors with small-scale structures which can
scatter particles: most likely these are Alfven waves.
I’m sure you remember these waves, from last term.
They are transverse waves, which are not compressive,
and which propagate (in the simplest case) along the
magnetic field. Thus, they can be thought of as prop-
agating wiggles in the field lines. Their dispersion re-
lation is ω = kvA, with vA = B/√4πρ. Charged
particles interact resonantly with Alfven waves. A par-
ticle moving along B at some velocity v sees a Doppler
shifted frequency ω′ = ω(1− v/vA) = ω − kv. Now,
the particle will interact with the fluctuating E field
of the wave; if the particle “stays in phase” with this
fluctuating wave, it will undergo a strong interaction.
But this happens if the Doppler shifted wave frequency
is close to the particle’s natural frequency, its gyrofre-
quency. That is, the interaction is strong when
ω − kv = ±Ω(γ) (8.14)
For relativistic particles, with v ≫ vA, this condition
is equivalent to an equality between the particles gyro-
radius and the wave’s wavelength:
rL(γ) = γmc2
eB≃ λres(γ) (8.15)
Numerically, note that particles with γ ∼ 103 in a field
B ∼ 1 µG have a gyroradius – and thus a resonant
Alfven wavelength – on the order of an AU.
If the Alfven waves are turbulent (say, just arising from
a turbulent background plasma, such as the ISM), the
picture above carries over directly; vm becomes vA,
and the concept of a scattering length L must be re-
placed with the wave energy density and some mea-
sure of the scattering cross section, similar to that used
above. This process is slow in many applications; it
has dE/dt ∝ v2A/c2 (and thus is classically a “second
order Fermi process”). In addition, because it relies on
the wave-particle resonance, a particle at energy γ only
sees waves at λres(γ) – which can be a small fraction
of the total energy in the turbulence.
8.6.3 Turbulent shock acceleration
A faster version of turbulent acceleration ties the tur-
bulence to shock fronts. A shock must have v2 < v1(the post-shock velocity must be less that the pre-shock
46 Physics 426 Notes Spring 2016
velocity; right?). Thus, in the frame of the shock, the
flows on each side are converging. If the pre-shock
and post-shock plasma also carry Alfven turbulence, a
particle trapped in the shock region and bouncing back
and forth will undergo Fermi acceleration, but will suf-
fer only head-on collisions. With this picture, we can
specify the parameters in the above Fermi analysis. We
need two facts:
• The escape probability is the ratio of downstream to
upstream fluxes: η ≃ 4nrelv2/nrelv = 4v2/c.
• The energy gain of a particle per scattering turns out
to be
E′ = E[1 + ve1(v1 − v2) cos θe1/c
2]
[1 + ve2(v1 − v2) cos θe2/c2](8.16)
if ve1, ve2, θe1 and θe2 are the velocities and angles of
the particle (electron, say), before and after it is scat-
tered. After p bounces, we have
lnEp
Eo≃ 4
3p(v1 − v2)
c(8.17)
With these, we can collect the algebra as above and
find the spectrum predicted by this model:
f(E) ∝ E−s; s =v1 + 2v2v1 − v2
(8.18)
Now, for a strong shock, v1 = 4v2, which predicts
s = 2; higher s values will be produced if the velocity
jump in the shock is < 4.
8.6.4 Energy limits for Alfven acceleration
What are the limits on particle energies that can be
scattered and accelerated by Alfven waves? The
particle-wave resonant condition (8.15) makes this
easy to answer. For the lower limit, we noted last
term that Alfven waves can only exist for frequencies
ω < Ωp = eB/mpc, the proton gyrofrequency. This
translates to a lower limit on the particles energies that
can “see” the waves (given in chapter 8 of P425 notes).
The highest particle energy which can possibly be ac-
celerated by Alfven waves is determined by the max-
imum wavelength that can exist in the system (which
can’t be any larger, clearly, than the physical size of
the system). In practice, however, particle acceleration
is limited by losses occuring at the same time as the
acceleration. If the losses are due to synchrotron radi-
ation, higher energy particles lose their energy more
rapidly (equation 8.4); this leads to a second upper
limit on the energy range of accelerated particles.
Key points
• Cosmic rays: in the galactic setting
• Relativistic electrons: indirectly “seen”
• First stage particle acceleration: where, what ener-
gies?
• Fermi acceleration, “classical”
• Alfven wave acceleration, shock and/or turbulent
Physics 426 Notes Spring 2016 47
9 Synchrotron radiation
A single particle, undergoing gyromotion in a mag-
netic field, is of course accelerated; and thus it will
radiate. When the particle is subrelativistic, this pro-
cess is called cyclotron radiation; when the particle is
relativistic, the process is called synchrotron radiation.
We have already studied bremsstrahlung, a fundamen-
tal continuous radiation mechanism which is important
for thermal astrophysical plasmas. Synchrotron radi-
ation is the other common and important continuous
emission process. It is imortant for magnetized astro-
physical plasmas which contain a significant compo-
nent of relativistic electrons (or, yes, positrons) – com-
monly called a “nonthermal” plasma.
9.1 Total power
We start with the total power radiated by the particle,
over all frequencies. Let the particle have acceleration
a, with components a‖ along its direction of motion
(v), and a⊥ across the direction of motion (NOT the
direction of the magnetic field, here). In the particle’s
rest frame, the total power radiated as a function of
time is
P (t) =2
3
e2
c3|a(t)|2 (9.1)
In the observer’s frame, this becomes
P (t) =2
3
e2
c3γ4(
a2⊥ + γ2a2‖
)
(9.2)
Now, for gyromotion, we have a‖ = 0 and a⊥ =eBcβ⊥/γmc. Let the particle have pitch angle θ – so
that β⊥ = β sin θ. Then, the net power seen by the
observer, per particle at energy γ and pitch angle θ, is
Psy =2
3
e4B2
m2c3γ2β2 sin2 θ (9.3)
You should note that this expression assumes γ ≫ 1.
From an ensemble of particles, with an isotropic dis-
tribution of pitch angles, we note that 〈sin2 θ〉 = 2/3;
thus the average single-particle power is
〈Psy〉 =4
9
e4B2
m2c3γ2β2 (9.4)
Finally, we note that (9.4) can be rewritten in terms
of the magnetic energy density, uB = B2/8π,
and the Thompson scattering cross section, σT =(8π/3)(e2/mec
2)2, as
〈Psy〉 =4
3cσTγ
2β2uB . (9.5)
9.2 Single particle spectrum
From the discussion of bremsstrahlung, recall: the
power as a function of time, P (t), reflects the time
dependence of the acceleration, a(t); the distribution
of this radiation over frequency – the spectrum – re-
flects the power (Fourier) spectrum of a2(t). In the
synchrotron case, the observed radiation varies with
time due to relativistic beaming effects, in addition to
the fundamental gyromotion. The power spectrum of
the observed P (t) turns out to contain power at fre-
quencies much higher than Ω/2π; this determines the
observed synchrotron emission spectrum.1
• First, guesstimate the answer. The important fact
here is that the radiation from a relativistic particle is
strongly beamed in the direction of the particle’s mo-
tion. A nonrelativistic particle radiates in a dipole pat-
tern, with the dipole axis aligned with the acceleration
a. However, for a relativistic particle, this dipole is
squeezed into a narrow forward cone, aligned with the
motion (v) and with opening angle ≃ 1/γ.
So, think about a relativistic particle in gyromotion.
You – as an observer – only see the particle’s radiation
once per orbit, when the particle’s velocity is within
1/γ of your line of sight. Say that the beam stays
within your line of sight for a time interval, ∆tobs.You will then see a very narrow pulse, of duration,
∆tobs, once every gyroperiod. From our experience
with Fourier transforms, we know that most of the ra-
diated power must appear at a frequency ∼ 1/∆tobs,
rather than just at Ω (as you might naively guess if you
didn’t think about the beaming effects).
We can estimate ∆tobs from geometry. Let the part of
the particle’s orbit for which you see the radiation have
length ∆s = a∆φ, where a is the radius of curvature
of the path. We can find a from the equation of motion:
∆v
∆t= v2
∆φ
∆s=
e
γmcvB sin θ (9.6)
so that a = ∆s/∆φ = γv/Ωo sin θ. Since ∆φ = 2/γ,
we have ∆s = 2a/γ = 2v/Ωo sin θ. Now, you see
a pulse of radiation; its emitted duration is ∆tem =∆s/v; and the duration you observed is shortened by
the light-travel time, so that ∆tobs = ∆tem − ∆s/c.Putting all of this together, we find
∆tobs =2
Ωo sin θ(1− β) ≃ 1
γ2Ωo sin θ(9.7)
1As usual, Ω = eB/γmc = Ωo/γ is the relativistic gyrofre-
quency.
48 Physics 426 Notes Spring 2016
where in the last step we have used 1/γ2 ≃ 2(1 − β),which is valid when γ ≫ 1.
Thus – we find that the observed pulse has a duration ≃2π/γ2Ωo (converting to Hz); we expect this to be the
highest frequency at which there is significant radiated
power.
• Now, do the math. When the calculation is done
more formally, the characteristic frequency for syn-
chrotron radiation turns out to be
νc =3
4πγ2
eB
mcsin θ (9.8)
When the power spectrum is evaluated, we find that the
radiation can be separated into components which are
linearly polarized along and across the magnetic field,
as seen projected on the sky. These are called P‖ and
P⊥. Considering the strong forward beaming effect,
we would expect the component polarized at right an-
gles to the field to dominate (why? Is this clear to you?)
The form of the spectrum comes out to be
P⊥(ν,E) =
√3
2
e3B sin θ
mc2
[
F
(
ν
νc
)
+G
(
ν
νc
)]
P‖(ν,E) =
√3
2
e3B sin θ
mc2
[
F
(
ν
νc
)
−G
(
ν
νc
)]
(9.9)
where
G(x) = xK2/3(x) ; F (x) = x
∫ ∞
xK5/3(x
′)dx′
and the K’s are Bessel functions. Since the F (x) and
G(x) functions behave similarly, this verifies that the
emissivity polarized transverse to the (projected) mag-
netic field is much stronger than the emissivity polar-
ized parallel to the field.
Clearly, the total power
P (ν,E) = P⊥(ν,E) + P‖(ν,E)
=√3e3B sin θ
mc2F
(
ν
νc
)
(9.10)
Both F (x) and G(x) peak at x ≃ 1. The asymptotic
behavior of F (x) is
F (x) → 4π√3Γ(1/3)
(x
2
)1/3x ≪ 1;
F (x) →(πx
2
)1/2e−x x ≫ 1
(9.11)
The function G(x) behaves similarly. Remembering
that x = ν/νc, this also verifies our guess: the radia-
tion peaks at νc and falls off exponentially above this.
Further, we note that the particle energy E = γmc2
enters the emissivity expressions (9.9) only through the
scaling of the argument, ν/νc.
9.3 Spectrum from a distribution of particle
energies
The next step is to find the spectrum radiated by a dis-
tribution of electron energies, f(E). Consider the total
emission, the sum of both polarized components, with
total power P (ν,E) = P‖(ν,E)+P⊥(ν,E). We have
for the total emissivity per volume,
jsy(ν) =1
4π
∫
P (ν,E)f(E)dEdΩ (9.12)
We will assume f(E) is isotropic, and integrate over
solid angle, in what follows.
Motivated by the observed cosmic ray spectrum, and
the photon spectrum seen in many polarized radio
sources, we choose the usual power-law particle spec-
trum,
f(E) = foE−s (9.13)
for the energy range Emin < E < Emax, where
Emax ≫ Emin is usually assumed. Note that there
must be an Emin; (9.13) diverges at low energies.
There may well be an Emax also – we’ll discuss that
later.
We put this DF into equation (9.12), and
write the single-particle power as P (ν,E) =PoBF (ν/coBE2). From the form of P (ν,E) and
the energy range chosen, note that we expect strong
emission in the frequency range
νmin < ν < νmax (9.14)
where νmin = νc(Emin) and νmax = νc(Emax).Putting this f(E) into (9.13), we get
jsy(ν) = PoBfo
∫ Emax
Emin
E−sF (ν/coBE2)dE
(9.15)
Now, by changing the variable of integration from Eto x = ν/coBE2, we end up with
jsy(ν) =PoB(s+1)/2foν
−(s−1)/2
×∫ xmax
xmin
x(s−3)/2F (x)dx(9.16)
Physics 426 Notes Spring 2016 49
Now, the integral in (9.16) can be evaluated numeri-
cally if xmin → 0 and xmax → ∞ (which is a reason-
able limit for the frequency range in (9.14); Pachol-
czyk, for instance, has numerical values. Thus, the
emissivity in (9.16) can be expressed numerically as
ǫsy(ν) = 4πjsy(ν)
= 1.18× 10−22a(s)foB(s+1)/2
(
ν
2c1
)−α
(9.17)
where c1 = 6.3× 1018 Hz, α = (s− 1)/2 is called the
spectral index, a(s) is an order-unity function (noting
the dependence of the x-integral in (9.16) on s), and
everything in (9.17) is in cgs units.
Equation (9.17) is good only for the frequency range
(9.14). Outside this range, the emissivity will be domi-
nated by that of particles at Emin (for ν < νmin), or of
particles at Emax (for ν > νmax). Thus, for ν < νmin,
we expect
low ν ′s : jsy(ν) ∝ ν1/3 ; (9.18)
and for ν > νmax, we expect
high ν ′s : jsy(ν) ∝ ν1/2e−ν/νmax . (9.19)
These limits may be repeated in the total spectrum
from a plasma – we’ll discuss this below.
9.4 Polarization
We noted that, since P⊥ > P‖, the radiation from a sin-
gle particle is linearly polarized. The fractional linear
polarization is ususally defined as
π(ν) =P⊥ − P‖
P⊥ + P‖(9.20)
For a single particle energy, π(ν) =G(ν/νc)/F (ν/νc). For a power-law distribution
of particle energies, both P⊥ and P‖ in (9.20) must be
integrated over particle energy; the result is
π =
∫
G(x)E−sdE∫
F (x)E−sdE=
s+ 1
s+ 7/3(9.21)
where we have taken x = ν/νc as above. In evalu-
ating the integrals, we have used the fact that the in-
tegrals∫∞0 xpF (x)dx and
∫∞0 xpG(x)dx can be ex-
pressed in closed form in terms of gamma functions.
Thus, a source with s ≃ 2− 3 will have ∼ 70% polar-
ization.
9.5 Synchrotron self-absorption
This is of course the inverse process – in which a free
electron in a magnetic field can absorb a photon. We
will treat this by relating the absorption probability
to the emissivity, taking stimulated emission into ac-
count, using a powerful statistical method developed
by Einstein, called Einstein coefficients.2 In order to
do this, we will represent the free electron energy state
as a discrete state in a continuum (and after all, even the
free electron phase space is quantized, remember), so
that the absorption is a transition between a lower state,
with energy E − hν, and momentum p(E − hν), and
an upper state with energy E and momentum p(E).
Referring to the Appendix, we see that the absorption
coefficient, at frequency ν, can be written in terms of
the populations of all pairs of upper and lower elec-
tronic states which are separated by hν of energy:
κν =hν
4π
∑
E
[N(E − hν)B12 −N(E)B21] (9.22)
where N(E) ≃ f(E)dE is something like, “the num-
ber of electrons in the Eth state”, if f(E) is the elec-
tron distribution function. Now, we note several facts:
• B12 = B21 for free electrons, which have the same
degeneracy factors g1 = g2 for the upper and lower
states;
• A21 = (2hν3/c2)B21;
• Psy(ν,E) = hνA21 relates the single particle emis-
sivity to the A21 coefficient;
Now, we switch back from describing discrete electron
states to a continuous picture:
∑
E
n(E) →∫
f(E)dE →∫
f(p)d3p
We can thus rewrite (9.22) as
κν =c2
8πhν3
∫
f [p(E − hν)− f [p(E)]
× P [ν,E(p)]d3p
(9.23)
Now, we can use the fact that the photon energy should
be small compared to the electron energy, hν ≪ E,
2We haven’t seen these formally; in case you haven’t seen them
in another class, I’m putting the important discussion in the Ap-
pendix to this chapter.
50 Physics 426 Notes Spring 2016
and expand the difference inside the braces in the inte-
grand:
f [p(E − hν)]− f [p(E)] ≃ hν
c
df(p)
dp(9.24)
Finally, since we are still assuming an isotropic particle
distribution, we can go back to energy space by noting
that d3p = 4πp2dp = 4πE2c−3dE (where the lat-
ter assumes the particles are highly relativistic). This
gives us a general expression for the synchrotron ab-
sorption from a distribution of electrons,
κν = − c2
8πν2
∫
P (ν,E)E2 d
dE
[
f(E)
E2
]
dE (9.25)
Equation (9.25) is still general, except for the assump-
tion of an isotropic particle distribution. Now, if we
specify f(E) to be the usual power law, and use the
same variable transform as in (9.16), we find the syn-
chrotron self-absorption from a power-law electron
distribution:
κsy(ν) =(s+ 2)c2
8πPofoc
(s+8)/2o
×B(s+2)/2ν−(s+4)/2
∫
x−(s+1)/2F (x)dx
(9.26)
The x-integral, again, can be evaluated as a function
of s; note, s also appears in the exponents of other pa-
rameters in (9.27). If we pick s = 2.5 as typical, we
can evaluate the constants in κsy(ν) (using cgs units,
still!):
κsy(ν) ≃ 8× 10−40foB(s+2)/2
(
ν
c1
)−(s+4)/2
(9.27)
9.6 Total synchrotron spectrum
Finally, we want to consider overall shape of the pho-
ton spectrum seen from a synchrotron source. When
we asked this question for a thermal source (such as
bremmstrahlung), we only had to deal with the radia-
tive transfer. Here, we must also deal with the underly-
ing electron spectrum – as we cannot assume it’s ther-
mal.
First, consider the effect of self-absorption on the
emergent spectrum from a synchrotron source. We re-
call the solution to the transfer equation:
Iν = Sν
(
1− e−τν)
(9.28)
where Sν = jν/κν and τν =∫
κνdx, integrated
through the source. This has the limiting solutions,
Iν → jνx for τν ≪ 1 (corresponding to high frequen-
cies for the synchrotron case), and Iν → Sν (corre-
sponding to low frequencies). Further, we note that
Sν ∝ ν5/2 from (9.17) and (9.27); since we have
assumed a non-Maxwellian electron distribution, we
should expect our source function not to be the low-
frequency limit of the black body spectrum.
The total spectrum from a source which has τν = 1at some observed frequency will thus have a low-
frequency range,
Iν ∝ ν5/2 (9.29)
and a high frequency range,
Iν ∝ ν−(s−1)/2 (9.30)
(this is the optically thin range).
This is not the full answer, however. We pointed out
earlier that the electron distribution must have a low-
energy cutoff, Emin (with critical frequency νmin =νc(Emin); and that it may also have a high-energy cut-
off, Emax (with critical frequency νmax = νc(Emax).If we consider these limits, but ignore transfer, then we
expect three frequency ranges. Thus, a purely optically
thin spectrum will have a low-frequency range,
Iν ∝ ν1/3 (9.31)
(why? compare the single-particle spectrum, 9.10 and
9.11). The source will have a mid-frequency range,
Iν ∝ ν−(s−1)/2 (9.32)
This source will also have a turnover at high frequen-
cies, due to a cutoff in the electron energy distribution
(say at γmax). The most likely spectral form is
Iν ∝ e−ν/νmax (9.33)
A variant of this can be obtained if the high-energy
electron distribution does not cut off abruptly, but more
slowly, as is the case in some models of electron aging.
Finally: what might be the cause of low-energy and
high-energy cutoffs? The high-energy cutoff is com-
monly assumed to be due to what’s called “spectral ag-
ing”. That is: from (9.4) or (9.5), recall that the single
particle power goes as P ∝ E2. Thus, if we start with
a power law electron distribution (as in 9.13), the high-
est energy particles will lose energy the fastest. This
Physics 426 Notes Spring 2016 51
leads to a truncation of the electron spectrum, at ener-
gies whose synchrotron lifetime equals the age of the
source. The low-energy cutoff is harder to specify. It
is likely to be due to the fundamental particle acceler-
ation mechanism. We discussed this in Chapter 8; it
may be that the low-γ limit of the resonance between
Alfven waves and accelerated particles produces this
Emin.
References
This comes mostly from my own notes. More details
can be found in
• Pacholczyk, Radio Astrophysics
• Rybicki & Lightmann, Radiative Processes in Astro-
physics
Key points
• Single particle synchrotron power and spectrum;
• Synchrotron emission from a power-law particle DF;
• Synchrotron self-absorption;
• Total synchrotron spectrum: high and low ν cutoffs.
Appendix: Einstein coefficients
I’m taking this directly from Rybicki & Lightman. Re-
member Kirchoff’s law, jν = κνBν (which we saw
back in radiative transfer). This relates emission to ab-
sorption for a thermal emitter; clearly there must be
some relation between emission and absorption at the
microscopic level (described by quantum mechanics).
To find it, consider transitions between two discrete en-
ergy levels of an atom, the first with energy E and sta-
tistical weight g1, the second with energy E+hνo and
statistical weight g2. There are three possible radiative
transitions between them:
• Spontaneous emission occurs when the upper level
spontaneously emits a photon; this occurs whether or
not the atom sits in an external radiation field. We
define A21 as the probability per unit time (sec−1) of
this happening (in principle A21 could be calculated
from quantum physics if we knew all the details of the
atom).
• Absorption occurs in the presence of photons of en-
ergy hνo. We define another coefficient, B12, such that
B12J is the probability per time for absorption. An im-
portant detail here: the spectral line associated with the
transition has some finite width, δν, about νo (due to
energy level uncertainty, doppler and collisional broad-
ening, etc). If φ(ν) is the shape of this spectral line,
we define J =∫
Jνφ(ν)dν as the weighted-mean in-
tensity over the line. Note that, if Jν varies slowly over
the line, φ(ν) acts like a delta function.
• Stimulated emission also occurs if there are photons
of energy hνo around. This wasn’t expected classi-
cally. Einstein found that it was needed in order to de-
rive Planck’s law; we now know it’s why masers mase
and lasers lase. We define a third coefficient, B21, so
that JB21 is the probability per time of a stimulated
emission event.
The game, then, is to use thermodynamic equilibrium
results to find relations between A21, B21 and B12. We
know three such results:
(i). In TE, the number of radiative transitions into state
1 must equal the number of transitions out of state 1.
If n1 and n2 are the number of atoms in the two states,
we know n1B12J = n2A21 + n2B21J .
(ii). In TE the ratio of level populations is given by
n2/n1 = (g2/g1)e−hνo/KT .
(iii). In TE the photon field is given by the Planck func-
tion: J = Bν (for ν = νo).
These three facts are enough: we can solve the equa-
tions to show that the Einstein coefficients must satisfy
g1B12 = g2B21 ; A21 =2hν3
c2B21 (9.34)
That’s the main result. We can – in principle – use
quantum physics to determine A21 for any given tran-
sition; then (9.34) tells us what the B’s must be.
The Einstein coefficients can also be used to determine
the emission and absorption coefficients. Going back
to the definitions (chapter 3), we can build
jν =hνo4π
n2A21φ(ν) (9.35)
and
κν =hνo4π
(n1B12 − n2B21)φ(ν). (9.36)
52 Physics 426 Notes Spring 2016
10 Pair plasmas in Astrophysics
Electron-positron pairs can also be important in the
physics of high energy plasmas and their interaction
with radiation. To set the stage: recall that the elec-
tron rest mass energy is 511 keV (or a temperature of
6× 109 K).
NOTATION: in this chapter, ǫ = hν/mec2 is the nor-
malized photon energy (relative to the electron rest
mass). The normalized lepton energy is γ = E/mec2,
as usual. Watch out: γ is also the “reaction notation”
for a photon (as in eqn. 10.1).
10.1 Pair annihilation
Free positrons will, of course, annihilate on electrons
(or any other form of “regular” matter). The most com-
mon decay is
e+ + e− → γ + γ (10.1)
The annihilation cross section, measured in the center
of momentum (CM) frame, is
σe+e− =πr2oγ + 1
[
γ2 + 4γ + 1
γ2 − 1ln(
γ +√
γ2 − 1)
− γ + 3√
γ2 − 1
]
(10.2)
Here, γ is the lepton energy (in the CM frame), and
ro is the classical electron radius, defined as ro =e2/mec
2 (in cgs). The Thomson cross section can be
written σT = 8πr2o/3.
The cross section has two useful limits. In the low-
energy case, β ≪ 1, the cross section becomes
σe+e− ≃ 1
βπr2o (10.3)
This shows that the annihilation probability is very
high for electrons nearly at rest. The decay produces
two photons very close to the rest energy of the lep-
tons: hν ∼ mec2, or ǫ = hν/mec
2 ≃ 1. This is the
annihilation line. In the high-energy case, γ ≫ 1, the
cross section becomes
σe+e− ≃ πr2oγ
(ln 2γ − 1) (10.4)
The decay at high energies still produces two photons,
but the photons have a much broader energy spread.
The annihilation line becomes a broad annihilation
spectrum.
It is worth remembering that the electrons and
positrons will undergo all of the usual plasma interac-
tions and radiation, during the time that they exist (be-
fore they annihilate). They can support plasma waves,
emit synchrotron radiation, and all of the other pro-
cesses we have studied.
10.2 Pair creation
Electron-positron pairs can be created in a wide variety
of nuclear and electromagnetic interactions. I summa-
rize the three mechanisms that seem most important
astrophysically.
10.2.1 Two-photon pair production
Classical physics says that EM radiation is linear; su-
perposition is allowed, without affecting either incom-
ing signal. This is not always true, however. We learn
from quantum physics that two photons can interact to
produce pairs, or other particles, if their energy is high
enough. The process of interest here is
γ + γ → e+ + e− (10.5)
It can occur if the total incoming photon energy, in
the rest frame, is at least as large as two electron rest
masses. In the lab frame, this translates to ǫ1ǫ2 ≥ 1(I’m keeping photon energies normalized to mec
2, to
shorten the notation. If you don’t accept this limit, do
the Lorentz transform for yourself, to check!). The
kinematics are simple in the CM frame: the two in-
coming photons must each have the same energy, and
likewise for the two created leptons. In the CM frame,
let β be the lepton velocity, and γ = (1 − β2)−1/2 be
its energy. Energetics require β2 = 1 − 1/ǫ1ǫ2. The
cross section for this process is
σγγ =πr2o2γ2
[
β(β2 − 2) + (3− β4) ln
(
1 + β
1− β
)]
(10.6)
This process can have an important impact on a lumi-
nous high-energy photon source (such as a gamma-ray
burster, or a very hot accretion disk). Consider the op-
tical depth of a gamma ray, at energy ǫ. It can react
with any photon that has energy above 1/ǫ. If L is the
luminosity of the source above this energy cutoff, then
the optical depth of the γ ray photon1 is
τγγ =Lσγγ
4πRǫmec3(10.7)
1Can you derive this?
Physics 426 Notes Spring 2016 53
This shows that the source can be opaque to its own
radiation, if it has high luminosity and small size. Such
a source is called “compact”.
It is also possible for pairs to be produced in photon-
proton or photon-electron interactions: γ + p → p +e++e−, and γ+e → e+e++e−. These reactions have
smaller cross sections than γγ pair production (smaller
by about the fine structure constant), and seem to be
less important astrophysically.
10.2.2 Pair production in pion decay
Another source of pair production is from pion decay.
Pi and mu mesons are unstable particles; once created,
they decay rapidly. The common decay chains are
π± → µ± + νµ/νµ
πo → γ + γ
µ± → e± + νe/νe + νµ/νµ
(10.8)
Thus, neutral pions decay simply to γ-ray photons.
Charged pions, however, decay to muons which in turn
decay to leptons.
So: how are pions made astrophysically? We have seen
one mechanism, the interation of cosmic-ray protons
with the microwave background:
γ + p → p+ π + · · · (10.9)
This predicts that there should be some free pions, de-
caying to muons and leptons, everywhere in space.
In addition, particle-particle reactons can make pions.
The most common (there are many more I’m not list-
ing. . .) are proton-proton collisions:
p+ p → p+ n+ π+
→ p+ p+ πo
→ d+ π+
(10.10)
The velocity-weighted cross section for π production
in a thermal pp reaction is
〈σppv〉 ≃ 4× 10−16 (lnT12) cm3s−1 (10.11)
if T12 is the proton temperature in units of 1012 K.
This form is good for temperatures above ∼ 100 MeV;
at lower temperatures the cross section drops rapidly.
One application of this process is to hot accretion
disks. Some models predict that the inner part of the
disks will be hot enough for pion production to take
place.
10.3 Magnetic pair production
Relativistic kinematics tells us that a free photon, in
vacuum, cannot create a massive particle (or particles);
the process cannot conserve energy and momentum.
Single-photon pair production is possible, however, in
the presence of a strong magnetic field. This process
requires high photon energies (in order to create the
lepton rest masses), and also high magnetic fields. The
field must be close to the so-called critical field, given
by
heBcrit
mec= mec
2 ; Bcrit ≃ 4.4 × 1013 G (10.12)
Energetics require that the photon satisfy
ǫ sin θ ≥ 2 (10.13)
if θ is the angle between the photon’s wavevector and
the magnetic field. The probability of this process oc-
curing is given in terms of an attenuation coefficient,
κ(χ) ≃ 1.5α
λc
B
Bcrite−4/3χ ; χ =
ǫ
2
B
Bcritsin θ
(10.14)
Here, α = e2/hc is the fine structure constant, and
λc = h/mec is the Compton wavelength of the elec-
tron. This quantity κ is essentially an absorption coef-
ficient: the “opacity” the photon sees is τ =∫
κ(χ)dx.
Magnetic pair production is thought to be important in
pulsars. You recall that these are strongly magnetized
neutron stars. The high electric fields induced by the
rapid rotation and strong B field pull charges off the
star’s surface and accelerate them to very high ener-
gies. These charges then emit γ rays, either by curva-
ture radiation (related to synchrotron emission) or by
inverse Compton scattering of thermal X-rays emitted
by the star. These γ rays can then pair produce. The
pairs themselves emit more γ rays, probably through
synchrotron radiation. These secondary γ rays then
make more pairs, which emit more photons . . . and
a pair cascade develops. The resultant dense pair at-
mosphere is thought to be the source of the coherent
pulsar radiation.
A very similar process may occur close to a massive
black hole in the center of an active galaxy. While
these models have not been as extensively developed
as pulsar models, the basic ingredients are the same.
The black hole, or the accretion disk feeding it, are al-
most certainly magnetized. Rotation will induce strong
54 Physics 426 Notes Spring 2016
electric fields, which can accelerate charges to high en-
ergies. In addition, the accretion flow is probably a
strong γ ray source itself. Thus, both two-photon and
single-photon pair cascades are possible in this setting.
Such a pair plasma may be the initial content of the
directed, relativistic radio jets created by the nuclear
black holes.
Key points
• Pair annihiliation: what it is, the magnitude of the
cross section.
• “Particle” pair production, two types, what it is.
• Magnetic pair production, what it is, its “probabil-
ity”.
Physics 426 Notes Spring 2016 55
11 (Inverse) Compton scattering
We have one more radiation mechanism to cover,
which is particularly relevant to compact objects and
relativistic (synchrotron) plasmas.
NOTATION: in this chapter ǫ = hν is the photon en-
ergy (in physical units, such as erg) ... sorry for the
switch, folks, it’s the standard notation.
11.1 Basic Tools
We need to use several different things here.
11.1.1 One event seen in the ERF
To start, go back to simple Compton scattering, as you
saw it in modern physics. Work in a frame where the
electron is at rest,1 and hit it with an incoming photon,
of energy ǫ′. The photon scatters through an angle θ′1,
and to an energy ǫ′1, with
ǫ′1 =ǫ′
1 + (ǫ′/mc2)(1− cos θ′1)(11.1)
Clearly, the electron gains momentum and energy after
the scattering, and the photon loses energy, ǫ′1 < ǫ′.
’ε
ε
θ
p
1= h ν ’’
’
’
1
Figure 11.1 The geometry of compton scattering, in the
electron rest frame (ERF). The photon comes in with ǫ′ =hν′, and leaves at angle θ′
1with ǫ′
1; the electron gains mo-
mentum p.
11.1.2 Cross sections
In order to get the Compton-scattered spectrum, we’ll
need to know the probability that the electron scatters
into angle θ′1 in the ERF, then average over all angles.
The probability comes from quantum electrodynamics,
and is called the Klein-Nishina formula:
dσ
dΩ=
r2o2
(
ǫ′1ǫ′
)2( ǫ′
ǫ′1+
ǫ′1ǫ′
− sin2 θ′1
)
(11.2)
where ro = e2/mec2 is the classical electron radius,
and dΩ is the differential solid angle for the scattering.
1This observing frame is called the Electron Rest Frame, ERF.
If we want the total scattering cross section, still in the
ERF, we integrate (11.2) over dΩ:
σ =
∫
dσ
dΩdΩ
The general result is a long expression, and the limits
are more useful. For x = ǫ/mc2:
x ≪ 1 : σ ≃ σT ;
x ≫ 1 : σ ≃ 3
8σT
1
x
(
ln 2x+1
2
)
(11.3)
(recalling that σT = 8πr2o/3). Thus, for low photon
energies, the scattering cross section is just σT ; for
high photon energies, it’s more complicated. Note,
finally, that these relations hold in the ERF. If we’re
looking at a scattering event in the lab frame, we know
that the photon energy is ǫ′ ≃ γǫ; thus the transition
energy in (11.3), namely x = 1, becomes γǫ = mc2.
11.1.3 Remember your relativity
Here’s a review if you need it.
• Lorentz transforms. Because our next step will be
transforming between the ERF and the lab frame, we’ll
need to remember some Lorentz transforms. Figure
11.2 has the geometry. You remember that x position
(along the motions) and time transform as
x′ = γ(x− βct) ; t′ = γ(t− βx/c) (11.4)
and, of course, the inverse transforms are just
x = γ(x′ + βct′) ; t = γ(t′ + βx′/c) (11.5)
(Think about which way the “frame” is moving, and
you can tell what to do about the + and − signs). You
also probably remember that the coordinates y and ztransverse to the motion don’t change.
But now: several other important physical quantities
transform just the same as position x and time do –
these are called 4-vectors. For a particle (massive or
massless), momentum p and energy E are a 4-vector;
so are wavenumber k and frequency ω if we’re work-
ing with a wave. We therefore have (keeping track of
units)
p′x = γ(px − βE/c) ; E′ = γ(E − βpxc) (11.6)
and
k′x = γ(kx − βω/c) ; ω′ = γ(ω − βkxc) (11.7)
56 Physics 426 Notes Spring 2016
c
x
y
z
x’
y’
z’
v= β
Figure 11.2 The geometry of our Lorentz transforms.
The primed frame is moving at v = βc along the x-axis;
the unprimed frame is the “lab”.
(and their respective inverses). Note that angle trans-
forms can be found from the components of p or k,
as needed. If we apply these results to photons, which
have E = pc and ω = ck, they collapse to one trans-
form,
ω′ = ωγ(1− β cos θ) (11.8)
where cos θ = kx/k selects out the k component along
the direction of motion. We’ll use this in a few min-
utes.
• Invariants. Some important quantities can be shown
to be invariant – that is, to have the same value mea-
sured in either the lab frame or the moving frame. I’m
not going to prove these here – you can look at Ry-
bicki & Lightmann if you’re curious. We’ll need two
important facts:
• We’re interested in the power radiated (say by an ac-
celerated particle). This is one invariant:
dE
dt=
dE′
dt′(11.9)
• We’ll want to work with the photon spectrum; let
n(ǫ)dǫ be the number of photons “at ǫ”. It turns out
that the ratio
n(ǫ)dǫ
ǫ=
n′(ǫ′)dǫ′
ǫ′(11.10)
is also an invariant.
11.2 Scattering as seen in the lab
What then is “inverse” compton scattering? If the elec-
tron is moving, and has more (kinetic) energy than the
photon, the photon will tend to gain energy in the col-
lision (at the expense of the electron). The only dif-
ference between this and the simple version in (11.1)
is the frame from which one views the collision ... but
the moving-electron case is traditionally called Inverse
Compton Scattering (ICS).
Thus, think of a situation where the photon has energy
ǫ = hν and the electron has energy γmec2 before the
scattering. We can analyze this by doing a Lorentz
transform to a frame moving with the (γ, β) of the
electron – the electron rest frame (ERF), and let it be
the primed frame – in that frame (11.1) applies. In the
ERF, we know the electron has β′ = 0 (by definition!),
and from (11.8), the photon has ǫ′ = γǫ(1 − β cos θ),where θ is the angle between the photon and electron
momenta in the lab. Let the scattering take place, then
transform back to the lab; the new photon energy in the
lab is
ǫ1 = ǫ′1γ(1 + β cos θ′1) (11.11)
(remember that ǫ′, ǫ′1 are related by 11.1). Thus, for a
given electron and photon energy before the scattering,
ǫ′ depends only on the scattering angle in the ERF.
11.2.1 Single particle radiation
The simplest application of this is to find the power
emitted by an electron exposed to some photon field.
By far the easiest way is to work in the ERF – where
the scattering is simple – and transform to and from the
lab as needed. We’ll use basic Lorentz transforms, and
the invariants (described above, in 11.9 and 11.10).
With this approach, it’s not hard to find the total power
scattered by our electron sitting in the radiation field
n(ǫ). In the ERF, that power is
dE′
dt′= cσT
∫
n(ǫ′)ǫ′1(ǫ′, θ′1)dǫ
′ (11.12)
where we’re assuming ǫ′1 from (11.1). The integrals
here are over the photon spectrum. Because of the in-
variants, it’s easy to write this in the lab frame:
dE
dt= cσT
∫
(ǫ′)2n′dǫ′
ǫ′
= cσTγ2
∫
(1− β cos θ)2ǫn(ǫ)dǫ
(11.13)
where we’ve used the basic Doppler shift, ǫ′ = γǫ(1−β cos θ) in the last step.
Now: let the photon field be isotropic. The angle
average of (1 − β cos θ)2 is just 1 + β2/3; so the
Compton-scattered power, angle-averaged and assum-
ing an isotropic photon field, is
dE
dt= cσTγ
2
(
1 +1
3β2
)
urad (11.14)
Physics 426 Notes Spring 2016 57
because urad =∫
ǫn(ǫ)dǫ is the radiation density that
the electron sees. Finally, then, we want the energy lost
by the electron – its Inverse Compton power. That’s
just the scattered power minus the incoming power:
PIC =dE
dt− cσTurad =
4
3cσTγ
2β2urad (11.15)
(Does this form look familiar? Compare the relativistic
limit, β ≃ 1, to the expression for synchrotron power,
equation 9.5).
11.2.2 Single particle spectrum
Next, we want the spectrum of the scattered radia-
tion. Unlike our previous applications (synchrotron
and bremsstrahlung), we’re not talking about Fourier
analysis here. Rather, we know that a single scattering
event leads to a photon energy ǫ1, given by (11.11).
But ǫ1 depends only on the angles θ, θ′1. One of these
is the input condition (the angle between the incom-
ing photons and the electron’s motion). For the other,
we know the probability of scattering into that angle –
from dσ/dΩ, (11.2). Thus, to get the photon spectrum
– the probability of scattering giving an output ǫ′ – we
just can carry out the angle average of (11.11), using
(11.2).
Because the power radiated depends on the photon
field as well as the electron energy, we need to as-
sume something about the radiation field. If the radi-
ation is isotropic and monoenergetic, at photon energy
ǫo = hνo, we can call its intensity I(ǫ) = Foδ(ǫ− ǫo).The angle averaging then gives (after much algebra),
Pic(ǫ1) = 3σTFoFic(x) (11.16)
where we’ve defined
x =ǫ1
4γ2ǫo(11.17)
and the kernel function is
Fic(x) = x(
2x ln x+ x+ 1− 2x2)
(11.18)
Compare the single-particle synchrotron spectrum,
from (9.9) and (9.11): it is a narrow function which
peaks at ν ≃ νc. Similarly, the function Fic is a nar-
row function which peaks at ν ≃ 2γ2νo. Thus we have
a “characteristic” frequency for inverse Compton scat-
tering, just as we had for synchrotron.
11.3 Composite spectra
In our previous derivations, we integrated the single-
particle spectrum over the distribution of particle en-
ergies, to find the volume emissivity (we did this be-
fore, for bremsstrahlung and synchrotron). It’s more
complicated than the previous derivations, because (i)
we have to assume an input photon spectrum, and (ii)
we really have to worry about whether a photon scat-
ters once, or many times. Instead of doing the general
problem, I’ll just look at two simple limits.
11.3.1 Nonrelativistic electrons
In this case, each scattering leads to only a
very small change in the photon energy: ǫ′1 ≃ǫ1[
1− (ǫ′/mc2)(1 − cos θ′)]
. If we use (11.2), and
do angle averaging, we find that the average energy
gain per photon, scattering on electrons at temperature
T , isδǫ
ǫ≃(
4kT − ǫ
mc2
)
(11.19)
Thus, one scattering can be significant to the radiation
source (for instance the microwave background, scat-
tering as it passes through a cluster of galaxies); but it
makes little difference to the photon spectrum, as long
as kT ≪ mc2.
11.3.2 Relativistic electrons, single scattering
The ICS kernel in (11.16, 11.18) peaks at ǫ1 ≃ 2γ2ǫo– that comes from the mean scattered photon energy.
(It’s easy to show that 4γ2ǫo is the maximum possi-
ble scattered energy; the mean will be somewhat less
than this). In most astrophysical applications, the ICS
opacity through a source is low, so the photons only
scatter once (if at all). Thus IC scattering on electrons
at energy γ boosts low-frequency radiation by ∼ γ2.
The low frequency photons might be synchrotron pho-
tons from the electrons themselves (in which case this
is called “synchrotron self-compton” radiation, SSC);
or they might be microwave background photons.
11.3.3 Scattering from power-law electrons
Put a distribution of relativisitic electrons, n(γ) again,
in a photon field with spectrum F (ν ′).2 If the photons
are monoenergetic at νo, we have F (ν ′) = Foδ(ν′ −
2Notation alert: in this section I’m letting ν′ be the incoming
photon energy – so that ν can refer to the scattered photon energy,
which is our desired final result.
58 Physics 426 Notes Spring 2016
νo) (from above). Looking back to (11.16, 11.18), it
will be useful to rewrite the ICS power from a single
scattering by a relativistic electron as
P (ν; νo, γ) =3
4
σTγ2
Foν
νofic(x) (11.20)
where fic(x) = 2x ln x+x+1−2x2. If we replace the
monoenergetic photon field by a spectrum F (ν ′), we
can find the emergent (singly) scattered ICS spectrum
by integrating over the photon spectrum and the energy
distribution:
4πjic(ν) = 3πσT
∫ ∫
n(γ)1
γ2ν
ν ′F (ν ′)fic(x)dγdν
′
(11.21)
where x = ν/(4γ2ν ′), as before. Because this is simi-
lar to the integrand we encountered in our synchrotron
analysis, we can use a similar variable transform. Let xand ν be the integration variables now; for our power-
law electron distribution take
n(γ) = noγ−s
as usual. With this, the integral in (11.21) becomes
jic(ν) =3
42sσTnoν
−(s−1)/2
∫
F (ν ′)(ν ′)(s−1)/2dν ′
×∫
x(s−1)/2fic(x)dx
(11.22)
This looks horrible, yes; but we’re almost there. The
last integral, over x, is just a number, because fic(x)is a simple function. In addition, the first integral, over
ν ′, depends only on the photon spectrum. If we specify
F (ν ′) (for instance the black body spectrum of the mi-
crowave background; or maybe the synchrotron spec-
trum of the electrons themselves), this first integral can
be worked out. Thus, we have the important result: the
ICS spectrum from a power-law electron distribution,
for single scattering, obeys
jic(ν) ∝ noν−(s−1)/2 (11.23)
That is: if the electrons are a power law, the ICS spec-
trum is also a power law, with spectral index (s−1)/2.
This is, of course, the same spectral index as that for
synchrotron emission from the same electrons – only
we must remember that the ICS photons come out at
much higher energies.
References
Once again, this is mostly from my own notes; but you
can find more details in
• Rybicki & Lightmann.
Key points
• Compton scattering: what it is, physically, and what
“inverse” is.
• Single particle ICS, power and spectrum.
• ICS from thermal (nonrelativistic) electrons.
• ICS from relativistic, power-law electrons.
Physics 426 Notes Spring 2016 59
12 Pulsars: overview and some physics
Carroll & Ostlie present the basic picture in some de-
tail; another good reference is Longair’s High Energy
Astrophysics. In these notes I’ll be brief about the ba-
sics, and emphasize the physics inside the basic model,
as well as newer work on high-energy emission and
pulsar winds.
12.1 The basic picture
What can cause a star’s brightness to pulse as quickly
as 100 times a second? That was the immediate ques-
tion when pulsars were discovered. Stars with longer-
period variability are “beating” – that is undergoing
body-mode oscillations. But we know a fair bit about
normal modes in stars, and none can have so high a
frequency. So we must consider rotation; perhaps a
hot spot on the star rotates past our line of sight? This
was more promising ... but the most compact star then
known, a white dwarf, is too big. Remember the rota-
tion rate is limited by lightspeed at the star’s surface, as
well as by the stability (gravitational binding energy)
of the star. Only neutron stars – which were predicted
by theory but not yet proved to exist – could explain
such a short period.
12.1.1 The cartoon
Thus the basic picture was born: an isolated pulsar is
a rapidly rotating neutron star with a small, radio-loud
“hot spot”. But why is there a “hot spot”? Why does
the star’s rotation slow down? We think the star is
strongly magnetized. We expect this to follow if the
NS is made in the core collapse of a supernova; flux
freezing will lead to a very strong magnetc field in the
NS. But this can, in principle, answer both questions.
Two things cause the star to slow down. A rotating
magnetic field emits magnetic dipole radiation; by en-
ergy conservation this must lead to spindown. If the
star sits in vacuum this is the only energy loss. If it sits
in the ISM, however, there will also be some torque be-
tween the star and the ISM, probably mediated by the
magnetic field and/or the plasma outflow from the star
(as discussed below).
If the star is strongly magnetized and if the magnetic
dipole is set at an angle to the rotation axis, we also
have a ready cartoon for why it pulses, as follows.
Close to the star the magnetic field will be dipolar and
will rotate with the star, as will any plasma which is
tied to these dipolar field lines. However, at the light
cylinder radius, that is rLC = c/Ω if Ω is the rotation
rate, the plasma cannot corotate with the star. Magnetic
field lines which start close to the magnetic pole will
not be able to turn around (“close”) before they reach
rLC ; rather they must connect to the B field of the local
ISM. Plasma can therefore flow out along these open
field lines. If the B field is dipolar, field lines within an-
gle ∼ (R∗/rLC)1/2 of the magnetic axis will be open.
If this outflowing plasma can – somehow – make in-
tense radio emission, which is strongly beamed for-
ward along the open field lines, then we will see strong
radio pulses only for the fraction of the rotation period
when the beamed radiation intersects our line of sight.
Figure 2.1.1 Cartoon of the standard picture of a pulsar.
From Ransom & Condon, NRAO.
12.1.2 And some details
• Typical numbers for single pulsars. The star’s ra-
dius R∗ ∼ 10 km (from NS models). The light cylin-
der is typically at ∼ 1000R∗ (but this of course de-
pends on the rotation rate Ω). The open field line
region is a few degrees across at the star’s surface.
From the duty cycle of the radio pulses, and the as-
sumption that the radio-loud plasma fills the open field
line region, we infer the radio emission comes from
∼ 3− 30R∗ above the surface.
• Energetics. An isolated pulsar (one not in a binary
system) is living off its rotational energy, In terms of its
rotation rate Ω and the associated period P = 2π/Ω,
60 Physics 426 Notes Spring 2016
this is
Erot =1
2IΩ2 =
2π2I
P 2;
Erot = IΩΩ = −4π2IP
P 3
(12.1)
As you remember (yes?) from basic mechanics, I is
the moment of inertia; if the NS were a homogeneous
sphere, we would have I = 2MR2/5. Pulsar people
generaly take M >∼ 1M⊙, and R ∼ 10 km. These are
probably pretty good guesses, although details of the
as-yet-unknown equation of state in the star’s core can
matter here. The 2/5 factor in I is much less certain,
due to our ignorance of the internal state of the star
and the importance of general relativity in the star’s
structure. Many authors choose I ≃ 1045 (cgs) as a
“typical” value, and scale to I45. Collecting these es-
timates, we find a large range for the power source,
Erot ∼ 1032 − 1038 erg/s, for old to young pulsars.
• How strong is the B field? Remember we have
no direct measure; and that very few NS sit in vac-
uum. Everyone in the field ignores this latter, and
assumes that magnetic dipole radiation dominates the
spindown. To remind you: magnetic dipole radiation
comes from the time-dependence of the star’s magnetic
moment, m. If this time dependence comes from the
star’s rotation, we get for a star rotating at Ω and a mag-
netic axis oriented at α relative to the rotation axis,
Pmag dip =2|m|23c3
=2
3
(Ω2m)2
c3sin2 α (12.2)
Now, the magnetic moment is connected to the mag-
netic field at the star’s surface and magnetic pole by
m = B∗R3∗/2. But P and P can be carefully mea-
sured; so, by equating Erot (from 12.1) to Pmag dip
(12.2), we can “derive” the B field: B∗ ≃ 3.2 ×1019(PP )1/2. Putting in typical P ’s and P ’s for sin-
gle pulsars, we find B∗ ∼ 1011 − 1013 G is the ex-
pected range. Finally, note that things get interesting
when the field is close to the quantum field, defined by
heBcr/mec ∼ mec2: Bcr ∼ 4.4× 1013G.
12.2 Spin a magnetic field
If the basic picture – a rapidly rotating, strongly mag-
netized neutron star – is correct, then some striking
physics follows. If you spin a B field, you generate
an E field. Think about ∂B/∂t, let z be the rota-
tion axis,1 and remember that the rotation velocity is
1I can’t do bold Greek; so please pretend Ω is a vector, Ωz.
vrot = Ω × r. The E field generated, measured in the
inertial (“lab”) frame, is
Eco = −vrot ×B/c = −(Ω× r)×B/c (12.3)
(I’m calling this the “corotation field”, for reasons
explained below). The important issue for the local
physics is whether this E field is felt, at full strength,
by the plasma around the star, and how that plasma re-
sponds.
To think about the impact of this on the NS, consider
two scenarios. Both are still current in the field, and
both lead to strong E fields, and high particle energies,
in the open field line region.
12.2.1 Star in vacuum
Inside the star, we assume the matter and field corotate.
The free charge must arrange itself into regions of pos-
itive and negative charge, so that the resulting E field
just balances the v × B force. Thus, there must exist
a physical E field, equal to that in (12.3), so that the
net force measured in the corotating frame is zero. But
this physical field has a non-zero divergence, so must
be supported by a local charge density:
ρco =1
4π∇ · Eco ≃ −Ω ·B
2πc(12.4)
(the ≃ means I’m dropping terms ∼ vrot/c; thus this
is valid only well inside the light cylinder). Thus: the
matter inside the star must have a net charge density,
which is quadropolar: one sign towards the two poles,
the opposite sign towards the equator.
Outside of the star, the E field will be a vacuum so-
lution (by assumption, there are no charges outside),
determined by the charge distribution of the star. But
we know that from basic E&M, we can just solve
Laplace’s equation (∇2Φ = 0, if E = −∇Φ) in the
vacuum region, matching the potential and tangen-
tial field at the star’s surface. Those solutions can be
worked out, but I won’t bore you with the details. The
interesting part of the solution is strength of the field,
E ∼ BΩR∗/c, and the fact that the E field in the polar
cap region has a strong component parallel to B (call
it E‖). From this, we can infer that any free charges
in the region will be accelerated to high energies – at
least in the open field line region. It may be that these
strong fields can even pull charges from the surface of
the star (which is mostly but not totally neutrons) ... so
that the external vacuum may not last for long.
Physics 426 Notes Spring 2016 61
12.2.2 Filled magnetosphere
Consider the other alternative, that the region above
the star’s surface is filled with plasma. We know the
charge density needed if this plasma is to corotate with
the star; it’s just (12.4). Most authors assume that
the plasma within the closed field line region has just
this charge density, so that the free charge shields the
rotation-induced E field, giving the net E = 0 (mea-
sured in the corotating frame). With no net force, the
plasma in the region will be static, again relative to the
corotating frame. In fact, consider the consequences
if the local plasma differed from ρco: there would be
a finite unshielded E field, and the free charges in the
plasma would move so as to cancel the field. Such a
situation would not be stable, nor likely to last for very
long. Thus corotating plasma is likely on closed field
lines.
The situation in the open field line region can be more
interesting, however. Charges leaving the star’s sur-
face must start at low velocity, and go through an ac-
celeration region in which they reach their final energy.
If the flow is steady, the charge density must vary in-
versely with the particle speed. In addition, particles
can leave the star along the open field lines, presum-
ably at high speeds. If the particle outflow carries a net
charge, a current is driven out from the star; as with
any electrical system, current density is sensitive to the
global circuit (return path, driving voltage and net re-
sistance). From these arguments we suspect that the
rotation-induced E is not fully shielded in at least part
of the open field line region. Any unshielded E‖ will
accelerate particles and drive the outflow/current.2
12.3 Radio emission and the pair cascade
Pulsars were discovered by their very intense, pulsed
radio emission. Many, many papers have been gener-
ated about observations and models of this emission.
However, it turns out to be only the small tail of the
dog: the total radio power is a very small fraction of
the spin-down power, Erot. In addition, we don’t know
what causes the radio emission. From the observed
very high brightness temperatures (TB ∼ 1035 might
be typical), we know the emission cannot be due to
2The real question is what the net, unshielded, potential drop
is, and thus particle energies are reached. Typical models suggest
Lorentz factors γ ∼ 106−107 are reached; but I wouldn’t suggest
overmuch confidence in this, the situation is complicated and still
not well understood.
any of the incoherent processes which apply elsewhere
in astrophysics. No plasma can be physically so hot
(why?? what would happen to the particles??) so TB
cannot be a physical temperature (as it would if it were
thermal emission); nor can this be synchrotron radia-
tion (remember the TB ∼ 1012K limit for synchrotron,
which we saw earlier). Thus we must be seeing a col-
lective or coherent emission mechanism; and these are
far from understood.
Despite lack of a solid model, a complex scenario has
evolved to describe where and how the radio emis-
sion is likely to occur. To start, let’s stay within the
open field line region, where we have just argued that
charges are accelerated to very high Lorentz factors.
There are two ways in which these particles can emit
very energetic photons.
• One way is curvature radiation. The charges must
follow the magnetic field lines (their gyroradii are tiny;
in fact their gyromotion is quantized). The field line
curvature makes the particles emit curvature radiation.
To understand this, go back to the arguments syn-
chrotron radiation; we can apply them here to curva-
ture emission. The characteristic photon frequency
is ∼ 3γ3c/4πρc, if ρc is the radius of curvature of
the field lines. The radiated power, integrated over
frequency, is ∼ e2cγ4/ρ2c (erg/s per particle). For
ρc ∼ 10 km, and the particle energies above, the cur-
vature emission photons come out in the γ-ray region,
in particular above the electron rest mass energy.
• Another alternative is inverse Compton scattering.
The pulsar itself is warm (we know they are thermal X-
ray sources). The primary charges can Compton scat-
ter the thermal photons to higher energy, again making
γ ray photons above mec2. [NOTE: in this situation we
cannot simply argue the scattered frequency ∼ γ2νo,
because of the special geometry, with the photons and
charges travelling nearly in the same direction. Do-
ing the calculation carefully does verify that some of
the scattered photons are hard enough to be interest-
ing, however.]
Either one of these mechanisms will generate pho-
tons which are energetically capable of one-photon
pair production, via γ + B → e+ + e− + B.
Recall that this mechanism has a low-B threshold,
hνB >∼ 0.1mec2Bcrit, which is very likely satisfied
in the high B fields near the pulsar polar cap. Once
created, the leptons probably have enough energy to
make more energetic photons (through synchrotron ra-
62 Physics 426 Notes Spring 2016
diation, most likely, also possibly further curvature and
IC emission). Thus a pair cascade occurs ... and is
thought to continue until most of the primary beam en-
ergy is converted to a dense pair plasma. Models sug-
gest typical Lorentz factors of the pairs ∼ 102 − 103.
Now what about the radio emission? As noted above,
we need some collective process. The story becomes
less clear at this point, as collective plasma emission is
not well understood. A general guess is that the pair
plasma is a necessary part of the picture. For instance,
plasma turbulence may be involved; the charges in
strong (large-amplitude),turbulent plasma waves can
show collective behavior and emit intense radio pulses.
If there is a residual E‖ in the pair-cascade region, it
will generate relative streaming of the electrons and
positrons; such streaming is known to lead to plasma
turbulence.
12.4 High altitudes and currents
The ideas in the discussion above have been around
for quite awhile, about as long as pulsars have been
known. In recent years, new telescopes (X and γ ray)
have expanded our picture of these stars and their in-
teraction with their immediate environment.
12.4.1 High energy emission
Pulsars are now commonly detected in X-rays and γ-
rays. To date, about 30 pulsars have been detected
in X-rays and similar numbers in γ-rays (out to 100
MeV). These tend to be the young ones, which will
have the largest Erot, and thus (possibly) have the
strongest high-energy emission; thus we might guess
that most or all pulsars would show high-energy emis-
sion if we had instruments sensitive enough to see
them.
The high-energy emission is pulsed, but less narrowly
so than is the radio emission. Thus, either the high-
energy emission comes from a different location in the
star, or (if contiguous with the radio emission) it is less
strongly beamed. Many authors suggest this radiation
comes from higher altitudes than the radio emission,
possibly even from the light cylinder region.
For the stars well-measured up to now, we know that
the radio emission is a very small part of the total
power; most of the luminosity comes out at high ener-
gies, above an MeV. While uncertainties about distance
and beaming factor make absolute power estimates dif-
ficult, we think that the bolometric power may be com-
parable to the spindown power. That means that pul-
sars are quite efficient at converting rotation energy to
hard photons; and that the radio emission – the obser-
vation which first detected these stars – is but the small
tail on a much larger dog.
What type of radiation are we seeing? Normal, inco-
herent synchrotron (from highly relativistic particles in
the star’s strong B field) seems to work well for the
Xrays. Leptons in the pair cascade are created with fi-
nite pitch angles, and lose their energy to synchrotron
radiation, much of which comes out as Xrays. The γ-
rays are thought to be the leftovers of the pair cascade
process, hard photons which escape the star without
being turned into pairs.
12.4.2 The pulsar circuit?
Here’s an important piece of unanswered physics in
pulsar models: what does the current do? That is ...
the basic low-altitude model says that free charges are
accelerated away from the star’s surface. This is what
starts the cascade that leads to the radio emission. But
this is a current; the rotating star acts as a unipolar dy-
namo (as we discussed earlier). But the star can’t build
up a net charge (why?) – so the current must close
somehow. How this happens has been the topic of dis-
cussion ever since these stars were discovered.
Your author does not claim to know the answer ... but
here’s a speculation. We know that charges flow most
easily along magnetic field lines – and undergo very
little electrical resistance along the way. But because
the polar current starts out along the open field lines,
which must connect to the ISM, we also know the cur-
rent must move across field lines, in order to connect to
other field lines which return to the star. This is likely
to happen in the outer magnetosphere (rather like the
cross-field charge flow which leads to earth’s aurora).
Further, cross-field motion is likely to be more resis-
tive; this dissipated energy may come out as observable
radiation. Might this be connected to the high-energy
pulsed emission, or even to a high-altitude component
of the radio emission?
12.5 Winds and nebulae
Now, let’s move out to larger scales, past the light
cylinder. The standard pulsar model, above, predicts
a strong Poynting flux radiated out from the star, and
also an outflow of magnetized, relativistic particles.
One might expect this energy outflow to couple to nor-
Physics 426 Notes Spring 2016 63
mal matter (to mass-load, somehow, when it reaches
the ISM); will it drive a wind out from the star? In
the inner regions, at least, this will be a relativistic,
strongly magnetized outflow, with significant angular
momentum – so we should not expect it to be spher-
ical. Rather, it should come out perpendicular to the
rotation axis, even if it is initially driven by the plasma
outflow from the star’s magnetic poles.3
12.5.1 Pulsar winds
This idea has been around for quite awhile; thanks to
recent technology (mostly X-ray satellites) we are now
seeing evidence of these winds. The data are striking.
For older pulsars (those not currently within SNR),
we sometimes see structures in the nearby ISM which
are clearly bow shocks associated with the star’s high-
speed motion through the ISM. We infer that an unseen
wind from the pulsar creates the pressure balance that
leads to the observed bow shock. From the standoff
distance of the bow shock we can estimate the wind
energy, and compare it to standard models of the pul-
sar.
For young pulsars (those still within their SNR), recent
CHANDRA images directly reveal the outflow from
the pulsars (the Crab and Vela pulsars are the best ex-
amples here). These outflows are complex: they show
jets, which presumably come out along the star’s ro-
tation axis (this is the only symmetry axis in the sys-
tem), and equatorial winds, which probably arise from
the combined effects of the star’s strong magnetic field
and its rapid rotation.
12.5.2 Pulsar wind nebulae
The wind coming out from the pulsar is (we think) still
highly relativistic, with charged particles moving al-
most exactly along the magnetic field lines. Thus it
should be nearly invisible (how would it radiate), ex-
cept for its dynamical effects (such as the bow shocks).
But what happens when it encounters the local ISM? If
that ISM is dense enough, the wind will shock down,
and the particles will be “thermalized” (they will gain a
significant component of energy transverse to the local
magnetic field). Thus, the shocked wind will become
visible, as a synchrotron source. This makes what is
called a pulsar wind nebula, PWN (we’ve already men-
tioned these in Chapter 7).
3Picture holding a rigid water hose at some angle, while you
spin around your vertical axis. Which way will the water go?
To date there are a several good examples of PWNe;
most of them are inside supernova remnants (which
provides the high ambient density/pressure that creates
the thermalizing shock in the wind). The Crab Nebula
is the most striking example of this phenomenon: the
shocked pulsar wind fills the nebula, and pushes a shell
of cooler matter outwards. This shell is made up of the
original SN ejecta and ISM which has been accumu-
lated along the way; it’s what we see as the optical and
radio nebula.
12.6 Magnetars and Anomalous pulsars
Here’s another new area. Based on some recent discov-
eries, of strong X/γ-ray flares, and/or strongly pulsed
X-ray emission, the picture described above is being
pushed in two ways. Two characteristics define these
unusual objects.
• They have extremely strong magnetic fields, B ∼1014G. This is well above the quantum critical
field, Bcrit ∼ 4×1013G – and in the regime where
all kinds of interesting physics happens (pho-
ton splitting, vacuum birefringence, ...). Such a
strong field may be too large to be due to simple
flux freezing in the SN collapse of the parent star;
various authors discuss dynamos taking place dur-
ing the SN collapse.
• They are not living on their rotational energy:
their strong flares have L ≫ Erot. Where, then,
does their energy come from? We think it’s mag-
netic – that the very strong B field inside the
star can occasionally break through the crust, in
a cross between a “starquake” and a very strong
“stellar flare”. Once this flux tube emerges into
the magnetosphere, reconnection will go, releas-
ing the magnetic energy as X-ray and γ-ray radi-
ation.
Because of the way these objects have been found, they
are often called SGRs (Soft Gamma Ray Repeaters)
or AXPs (Anomalous X-ray Pulsars), and appear to
be different objects; but the growing consensus is that
these are both observational variants of the magnetar
phenomenon. Although none of these objects were
initially found in radio, one or two have now been de-
tected to have “normal” radio pulses as well.
Your author doesn’t know just how magnetars fit into
the general pulsar picture, they are too new – but their
64 Physics 426 Notes Spring 2016
existence points out that the range of extreme condi-
tions which can exist on or around neutron stars seems
to be more diverse than we’ve understood up to now.
Key points
• The basic picture, a pulsar as a rotating magnetized
neutron star;
• Our cartoon of the pulsar magnetosphere: what is the
plasma doing?
• The pair cascade, how it fills the magnetosphere;
• How the pair-filled magnetosphere might make radio
and high-energy emission;
• Pulsar winds and nebulae.
• Magnetars.
Physics 426 Notes Spring 2016 65
13 Radio jets and radio galaxies
The idea of astrophysical jets first attracted interest as
a theoretical prediction. Double-lobed radio galaxies
(RGs) were detected in the first radio surveys (in the
1960’s?). The data available then showed two radio-
loud lobes (synchrotron sources, containing magne-
tized plasma and relativistic particles) on either side
of a central, elliptical galaxy. Energetics and lifetime
arguments soon found that the lobes were short-lived.1
Either we were seeing them at a very special time in
the life of the parent galaxy, or they were being re-
supplied with energy by an undetected pipeline: a ra-
dio jet. These jets were detected when the instruments
improved – this was one of the first successes of the
VLA. Models of active galactic nuclei – involving ac-
cretion onto a black hole – had to be amended to in-
clude the production of highly collimated, relativistic
plasma jets, if the AGN happened to live in an ellipti-
cal galaxy.
We’ve learned a lot since then. We now know that jets
are common on both small and large scales. On large
scales, the massive black holes in galactic nuclei can
produce jets while they are “active”. Radio galaxies
and (radio-loud) quasars, which live in elliptical galax-
ies, are the most dramatic examples. In these objects,
the galactic nucleus contains a bright, compact radio
core and a pc-scale, synchrotron-bright jet. The ra-
dio core is thought to be optically thick synchrotron
emission from the base of the jet. On larger scales, the
jets extend to at least several kpc, often farther; they
connect to larger lobes or tails which arise from the
interaction of the radio jet with the local extragalactic
plasma.
On smaller scales, we now know that many star-sized
galactic sources have relativistic jets. The most well-
studied (and well-imaged) are jets from accretion flows
in X-ray binary systems; SS433 is the prototype, and
a few dozen are know known (microquasars). In addi-
tion, a few pulsars/PWNe systems are found to have jet
outflows: the Crab and Vela pulsars are examples here.
Finally, it is now thought that Gamma-Ray Bursters
(GRBs) involve highly relativistic jets, and that core-
collapse supernovae produce short-lived jets during the
explosion. Also on stellar scales, but moving to less
1Equipartition calculations give you the minimum energy in a
radio source; divide this by its radio power, and you get a “lifetime”
– which is short compared to any plausible estimate of the source
age. Thus, you need energy resupply.
relativistic systems, we now know that the collimated
outflows are produced during part of the collapse of
a protostellar cloud to the final star; these are called
(protostellar jets).
While all of these jets are interesting, in these notes
I’ll focus on what I’m most familiar with, namely,
relativistic jets from AGN. My goal is to present an
overview of the observations, some basic physics, and
the current “cartoons” as to how RGs work.
13.1 Jets: the observational constraints
To start, what are the important properties of jets which
theory must accound for? In these notes, the discus-
sion is strongly skewed toward extragalactic jets (radio
galaxies and quasars), and relativistic galactic jets (mi-
croquasars), which is my personal interest and area of
experience. Much of the basic dynamics apply to all
jets, but some of the details and constraints are more
relevant to relativistic jets.
Some critical facts and problems are:
• Internal energy. Radio jets from AGN and micro-
quasars are dominantly synchrotron sources – thus we
know they are magnetized and internally relativistic.
Protostellar jets and some galactic jets are dominated
by cooler, thermal radiation: emission lines from ion-
ized gas, and even molecular emission.
• Collimation. These jets usually have very small
opening angles – no more than a few degrees – and of-
ten retain their direction and collimation for a distance
which is orders of magnitude larger than the scale on
which they were initially produced.
• Knots and bright spots. Jets are rarely smooth when
seen in high-quality images. Bright features are com-
mon. We don’t know just why these features appear,
but a mixture of shocks and strong-amplitude waves in
the flow are likely. Note also that the radiation mech-
anisms are strongly nonlinear amplifiers: a weak den-
sity or B field enhancement, for example, can cause a
strong enhancement of the emissivity.
• Speed. There is indirect evidence that many jets in
RGs are supersonic (relative to themselves): structures
are seen at their outer ends, where they run into the
ambient medium, that can be identified as shock fronts
within the jet. In addition, on pc scales, some jets (or
at least waves in the jets) are moving at relativistic
speeds; this is inferred from the apparent proper mo-
tion of knots in the jets, which can exceed lightspeed.
66 Physics 426 Notes Spring 2016
There’s no compelling evidence for relativistic motion
in kpc-scale jets; somehow the high-γ pc-scale flow
has slowed down by the time it reaches kpc scales.
• Plasma content. Protostellar jets appear to be nor-
mal ISM; an ion-electron mixture. Compact-object
and extragalactic jets are observed in synchrotron ra-
diation, and therefore contain relativistic electrons and
magnetic fields; we do not know if the electron charge
is balanced by ions or by positrons.
• Energization. In at least some extragalactic jets
there is a clear need for local re-acceleration of the
relativistic particles. The synchrotron lifetime of the
highest energy particles is significantly less that the
shortest possible travel time down the jet (jet length
/ lightspeed). Somehow, local fluid/plasma processes
must transfer energy from the bulk flow to the indi-
vidual particles; a mixture of shocks and turbulence is
probably the answer.
13.2 Some useful relativity
Two relativistic effects are important here.
13.2.1 Superluminal motion
The bright knots in many jets are observed to have ap-
parent proper motion above lightspeed. As you’ve seen
before (for instance in Caroll & Ostlie), this is a simple
consequence of light-travel times. If the jet is oriented
at angle θ to the line of sight, and has physical speed
βc, its apparent speed – as seen by the observer – is
βapp =β sin θ
1− β cos θ(13.1)
It’s easy to show that this can lead to βapp > 1 for
θ ≪ 1. Typical observed values are βapp ∼ a few;
you should be able to work out what (γ, θ) values are
needed for this.
13.2.2 Doppler beaming
A source of radiation is moving relativistically, at some
γ, at angle θ to your line of sight. One effect, which
you remember, is a Doppler shift. If the source emits
photons at ν ′, you observe photons at ν, where the
observed and emitted frequencies are related by ν ′ =γν(1 − β cos θ) (note, the angle θ is measured in the
observer’s frame; and θ → 0 means motion towards
the observer). This can be written,
ν ′ =ν
D ; D(θ) =1
γ(1− β cos θ)(13.2)
θ
ct
vt cos θ
vt s
in θ
Figure 13.1 Illustrating how a feature in a jet can appear
to be superluminal. The core source (small circle at left)
emits a “blob” (large circle) at time t = 0, at angle θ to the
line of sight; it also emits an EM signal directly towards us.
After time t, the blob has moved a distance vt, and emits
a second signal. The difference in arrival times between
the first and second signals is ct − vt cos θ; the apparent
separation of the core and blob is vt sin θ. Dividing the latter
by the former gives us the apparent velocity of the blob, as
in (13.1).
where D is called the Doppler factor. A second impor-
tant fact is how to connect the spectral intensity you
observe, Iν , to that emitted by the source, I ′ν′ . It turns
out that the two quantities are related by
Iνν3
=Iν′
(ν ′)3(13.3)
(cf. Rybicki & Lightmann for the derivation of this).
As an example, let our emitting soure have a power-
law synchrotron spectrum, say S′ν ∝ ν−α in the rest
frame. When you work out the details, an unresolved
blob is Doppler boosted by Sν(θ) = S′νD3+α; you
get 3 D’s from the frequency ratio in (13.3), and an-
other “α of a D” from the spectrum. Alternatively, a
piece of a resolved jet is boosted by Sν(θ) = S′νD2+α;
time/space contractions applied to the piece you’re re-
solving account for the change. Either way, because of
the strong dependence of D on θ, this result means that
an observer sees the radiation forward-beamed, into an
angle ∼ 1/γ. This makes a relativistic jet appear much
brighter if you see it end-on.
13.3 Some useful physics
In this section I store a smattering of physical argu-
ments we can make about jets.
13.3.1 Collimation
How to jets stay so well collimated? Why do they
not expand? If the jet were propagating in vacuum, it
would expand at its internal sound speed. But we ob-
Physics 426 Notes Spring 2016 67
serve jets which remain very collimated over very large
distances, with an opening angle only a small fraction
of a radian. This would require the flows to be very
cold internally, which is inconsistent with other evi-
dence that many of the jets are internally hot. Thus we
believe the jet is confined. The two possibilities are
• Confinement by external pressure. We know jets
propagate through external plasma – the ISM for mi-
croquasars, the extragalactic medium for jets from
AGN. The external plasma may provide enough pres-
sure to confine the jet.
• Self-confiment by magnetic fields. We’ve already
seen that a current generates an azimuthal magnetic
field, which can confine the plasma carrying the cur-
rent. If this applies to jets, the question is then, where
and how does the current return to the source?
Which of these two operates can, in principle, be
learned from observations; and a given jet may change
from being self-confined (close to its origins, say) to
being pressure confined (farther out).
13.3.2 Jet transport
Some simple models of jet/RG evolution are based on
the rate at which mass, momentum, and energy flow
down the jet. Looking back to last term (when we did
mass and momentum conservation .. right?), and ear-
lier this term (for energy conservation) we can use the
basic fluid equations, and/or simple common sense, to
write down the rates. Let the jet have radius rj , speed
vj , density ρj , and enthalpy hj = ej + pj/ρj (this
is a useful way to collect terms involving the internal
energy and pressure of the jet fluid). If everything is
subrelativistic, and the magnetic field can be ignored
for the moment, we have
mass flux : M = πr2jρjvj (13.4)
momentum flux : P = πr2jρjv2j
(
1 +hjc2
)
(13.5)
energy flux : E = πr2jρjvj
(
hj +1
2v2j
)
(13.6)
If t he flow is relativistic, but still ignoring B, these
become (remember v = βc and γ = (1− β2)−1/2:
mass flux : M = πr2jρjγjβjc (13.7)
momentum flux : P = πr2jγ2j β
2j ρj(
c2 + hj)
(13.8)
energy flux : E = πr2jγ2j βjcρj
(
c2 + hj)
(13.9)
If the flow is strongly magnetized, we need to include
the field energy in the bookkeeping; you remember
that electromagnetic energy is transported in a Poynt-
ing flux. The details of this are complex and more than
we need here; I’ll return to this general idea below.
13.4 Larger Scales: the Radio Galaxy
If a jet propagated into vacuum, we might imagine that
it would remain unchanged, carrying on forever. But
jets don’t live in vacuum; rather they propagate into the
surrounding plasma – which can be relatively dense in-
tracluster medium (ICM: if the parent galaxy lives in a
cluster), or the tenuous intergalactic medium (IGM; if
the parent galaxy lives in the “field”). So: when the jet
interacts with the surrounding medium2 the nature of
this interaction shapes the Radio Galaxies (RGs) that
we see. Observed RGs tend to fall into two morpho-
logical types3, suggesting two different physical situa-
tions.
13.4.1 Classical Double radio galaxies (FR II’s)
Radio galaxies classified as FR II’s – cartooned in Fig-
ure 13.2 – are identified by their symmetric, two-sided
“lobes” which have bright “hot spots” at their outer
edges. Often a narrow, well-collimated jet can be seen
propagating through the lobe and almost to the hot
spot. These tend to be the brightest ones – so that, even
though they are rare by number (only a few per cent of
the population), they have received the most attention.
Here’s the basic scenario for this type of source. Put
a massive black hole down in the center of a galaxy,
and turn it on. The jet propagates out, into the ICM; as
time goes on, the head of the jet will reach farther and
farther from the AGN. We can use simple conserva-
tion laws to estimate how this souce evolves with time.
The mass and energy flowing down the jet carry mo-
mentum; this momentum flux allows the end of the jet
to make its way into the external medium (with den-
sity ρx). Think about a simple ram pressure balance:
if the end of the jet advances at vD = dD/dt, it sees
a head-on ram pressure ρxv2D . Balancing this against
the momentum flux in the jet tells us what vD must be.
2I’ll refer to the ambient medium as the “ICM” for short ...
meaning ICM/IGM.3Jargon: two authors, Fanaroff & Riley, first pointed out this
duality – so RGs today are still usually called “FR I’s” (FR Type
I’s) or “FR II’s” (FR Type II’s).
68 Physics 426 Notes Spring 2016
D(t)
core
ambient
mediumlobe
jet
Figure 13.2 Cartoon of (half of) an FRII radio galaxy. Let
a jet have constant opening angle, propagating into an ex-
ternal medium (at density ρx, say). Because the length of
the jet does not grow as fast as the plasma speed within the
jet, the plasma must slow down (possibly shock down), and
move “sideways”, creating a larger “lobe” which surrounds
the jet. Because the shock compresses the jet plasma and B
field, and accelerates relativistic particles, we expect it to be
a bright synchrotron source – i.e.the “hot spot” seen in many
such sources.
If the jet is less dense than the external medium, then
vD will be less than the jet speed; if the jet is also su-
personic we expect a shock to form at this transition
point.
Two things should happen at this shock. First, the
jet plasma will be compressed and heated when it
shocks down. The higher plasma density, and higher
B field, will make the post-shock plasma a stronger
synchrotron source; if any particle acceleration takes
place at this shock, that will further enhance the syn-
chrotron power. Thus, the post-shock plasma should be
a localized “hot spot” – matching nicely with what we
observe in classical double RGs. Second, the shocked
plasma must go somewhere – we expect the high post-
shock pressure to “push it off to the side”. As the RG
grows, this shocked jet plasma will expand to fill a
“lobe” or “cocoon” which surrounds the jet. This also
matches what we see – in fact the lobes are the bright-
est, defining, parts of this type of RG.
Recent observational note: if the jet is strong enough,
we expect its advance speed, vD, to still be supersonic
relative to the ICM. If this holds, the advancing jet will
drive a bow shock in the ICM – such shocks have now
been seen in a few cases.
13.4.2 Tailed radio galaxies (FR I’s)
Radio galaxies classified as FR I’s – as cartooned
in Figure 13.3 – are characterized by long, diffuse-
looking “tails”. These tails begin close to the AGN
(sometimes you can see a well-collimated inner jet that
suddenly changes to a broader tail; other times you
can’t), and carry on – broadening gently – for up to
100s of kpc. Unlike FR II’s, which are brightest at their
outer hot spots, FR I tails tend to be brightest close
to the galaxy; the tails gradually become fainter going
away from the core. The physical end of the flow –
where it runs into the ICM – may or may not be visi-
ble.
core
jet
"tail"Figure 13.3 Cartoon of (half of) an FRI radio galaxy.
When the jet initially leaves the AGN, it is well collimated
(probably relativistic and supersonic as well, at least in some
cases). But it soon destabilizes – sometimes very suddenly,
as in this cartoon. The plasma flow carries on, away from
the AGN, but in a more disorganized way.
FR I’s are much more common than FR II’s; most RG’s
we know are FR I’s. But FR II’s have been better stud-
ied, partly because they are the bright ones (so were
the first ones to be well-observed), and partly because
they seem to involve simpler physics.
So, what is the physical picture for FR I’s? The cartoon
above, for FR II’s, relies on the jet remaining stable,
collimated, and (internally) supersonic.4
The situation will be different if the jet becomes unsta-
ble, uncollimated, or subsonic. In this case, the jet flow
not retain the “jet/lobe” morphology characteristic of
FRII’s. Instead, we expect the jet flow to be much more
sensitive to local conditions in the ICM. The flow can
be affected by local ICM “weather” (flows, turbulence
in the ICM); if the parent galaxy is moving rapidly
through the local ICM, we can see strongly bent radio
tails. If the flow is slow enough, and less dense than its
surroundings (which is almost always the case), it will
also be affected by buoyancy.
Clearly a wide range of radio-tail morphologies is
possible here, depending on local conditions (ICM
weather), and also the details of the jet flow. How-
ever, the range of possible flows should have one thing
in common: they are probably brighter synchrotron
sources closer to the AGN (whereas FRII’s are brighter
closer to their hot spots). We expect this because the
radio tail usually expands as it propagates – leading
4These conditions are related – it turns out that subsonic, or
subalfvenic, jets can easily be destabilized as they pass through a
surrounding plasma. (Think of a firehose flapping side to side if
the pressure gets high enough ...) Supersonic/superalfvenic jets,
on the other hand, tend to be relatively stable in the same situation.
Physics 426 Notes Spring 2016 69
to lower plasma density and B field – thus lower syn-
chrotron power. Synchrotron aging may also matter –
the ends of the tails can simply fade away as the rela-
tivistic electrons lose their energy.
Simple dynamical models (as we have for FR II’s) have
yet to be developed for FR I’s. While we expect the
basic momentum-conservation picture, from above, to
hold here, the added dynamical effects of the interac-
tion with the ICM make it harder to find simple models
of FR I evolution.
13.5 Unresolved issues
While much of the above picture has remained stable
for quite awhile, some issues and questions are getting
new attention.
13.5.1 What is the life cycle of a RG?
Two questions come to mind here.
First, what is the duty cycle of a radio-loud AGN? If ∼10% of bright ellipticals host a radio galaxy, does that
mean that the black hole in every such galaxy is active
only 10% of the time? Do AGN, or RG’s, alternate
between “on” and “off” phases?
Second, where are the old radio galaxies? The sim-
ple models described above predict that RG’s should
keep growing, and remain relatively bright synchrotron
sources (i.e., detectable), as long as the jet stays “on”.
If the jet turns off, the same models predict that the
RG’s synchrotron luminosity should slowly fade, as
the relativistic electrons lose their energy. Because the
synchrotron lifetime tends to be long for typical RG
conditions, this fading should be slow – we should ob-
serve a fair number of “old” RG’s (jetless, probably
steep radio spectrum).
But we don’t see what these models predict. We al-
most never see RGs that could be called “old”; they
all have currently-active jets. In addition, dynamical
models (such as you’ll see in the homework) generally
estimate RG ages only ∼ 100 Myr – much younger
than the age of the parent galaxy. So: why don’t we
see either older, jet-on sources, or old, jet-off sources?
13.5.2 How does a jet affect its environment?
This is a very active current area. Clearly the envi-
ronment affects the AGN; accretion from the local en-
vironment is what makes the massive black hole “ac-
tive”, and creates the jet. We know the jet transports
mass, momentum and energy out from the AGN: what
effect does this transport have on the environment?
If the jet/RG is well-coupled to its environment – and
that’s a big “if” – then the jet/RG can have a strong
effect. It carries enough energy to heat the local ICM
significantly. Two applications here:
•Cool cores in clusters of galaxies. In most clusters
(as we’ve seen in the homework), the plasma density
is low enough that the radiative cooling time (from
bremsstrahlung) is longer than the Hubble time. Thus,
most of the plasma in most clusters hasn’t cooled down
by much since the cluster first formed. In some clus-
ters, however, the central plasma is dense enough that
radiative cooling does matter; these “cool cores” must
have an ongoing heating mechanism. All of these
cool cores are observed to be centered on a currently-
active AGN and RG; do these RG heat the cluster core
enough to offset radiative losses?
•Feedback in galaxy formation. We now know that
a massive black hole sits at the heart of every galaxy.
We suspect that black hole formed, and was “active”,
at about the same time as the galaxy originally formed
(we’ll discuss this later, in chapter 15). So: did the
energy released by that actively accreting black hole
play an important feedback role in the galaxy forma-
tion process?
Both of the above questions are getting a lot of atten-
tion these days. In my opinion, the big, unanswered
physics question is, “how effectively does the jet/RG
couple to its environment?” That is: how much of its
energy does the radio jet/RG actually transfer to the
local ICM? The answer depends on the details of how
the RG grows and evolves. Does it simply do “pdV ”
work as it pushes the ICM/ISM out of its way? Does it
drive significant shocks or turbulence in the ICM/ISM
(which then dissipate and heat the ICM/ISM)? Does
the relativistic plasma mix effectively with the ICM
– thus releasing the relativistic plasma directly into
the ICM/ISM? None of these questions have been an-
swered yet.
13.6 How are jets made?
Finally, a few words about how this all starts.
It seems very likely that every accretion flow involves
a jet outflow. And, nearly all jets that we know about
are tied to accretion flows (the one exception might be
jets from single pulsars; we don’t yet know much about
them). However, we don’t yet have a clear and agreed-
70 Physics 426 Notes Spring 2016
upon picture of how jets get formed. Models out there
can be grouped into two broad categories, fluid-based
and MHD-based. I think the MHD models are most
likely to prove correct; but will include a brief discus-
sion of wind models as well.
13.6.1 Wind (fluid-based) models
These models are probably not the right answer, and
pretty much disregarded in recent work. However they
were the first type of model proposed; and the internal
physics may well still be useful in the more modern
MHD models, below. To think about wind models, re-
member the solar wind: it accelerates from a very slow
start, to supersonic speeds at large distances. It does
this smoothly, by passing through a “gravitational noz-
zle” at the sonic point. The wind is driven by its in-
ternal energy (and thus, ultimately, by whatever heats
it). Linear, one-dimensional flows can also undergo a
smooth transition, if the area of the confining channel
has a minimum at the sonic point. The first models
of astrophysical jets used this analogy – arguing that
the flow is accelerated by its internal energy, and that
some combination of channel geometry (provided per-
haps by the walls of a fat accretion disk?) and gravity
produce a high-speed, somewhat collimated outflow.
13.6.2 MHD models
In this discussion I’m partly following several recent
papers5 and addressing jet formation from magnetized
accretion flows around a black hole.
What might the magnetic field of an accretion flow
look like? Think about a field which threads the disk
plasma, but is also “tied” to the distant ISM, which
is rotating much more slowly than the accretion flow.
The accreting gas will draw the field with it, imparting
a φ component to the field. One might expect dissipa-
tive processes to balance the field growth induced by
the accretion, leading to a field shaped something like
an expanding helix. Plasma ejected from the disk (for
instance in a wind) will be both channeled and accel-
erated by such a field. Magnetic pressure gradients up
the rotation axis will accelerate plasma “up and out”
of the system; magnetic tension (from the helical field
lines) will exert a “hoop stress” and confine the plasma
as it goes.
5e.g. Meier, D. L. 2005, Astrophysics & Space Science, 300,
55; Meier, D. L. & Nakamura, M. 2006, ASP Conference Series,
350, 195
There are also a couple of variants worth mentioning.
• Poynting flux models. These describe the limit in
which the plasma density close to the source is small
compared to the field energy. As we’ve seen, a rapidly
rotating B field generates an E field (just as in pulsars).
There will very likely be a component of the Poynt-
ing flux, S ∝ E × B, along the rotation axis, which
will be a strong source of power lost from the accre-
tion system. In order to produce what we observe as
radio jets, these Poynting-flux jets must “mass-load”;
they must either pick up stray charges from the ambient
medium, or generate their own plasma through magne-
tized pair production. Such a process may well account
for the origin of relativistic jets close to a black hole;
they would then become visible as they gained mass.
• Penrose jets. Another class of models extracts rota-
tion energy from the black hole to power the jets. (I’m
inventing the name; the authors of two similar mod-
els are Blandford & Znajek, and Punsly & Coroniti).
Think back to rotating black holes. You recall that
frame dragging is important near the event horizion –
this is the azimuthal motion induced by the hole’s ro-
tation. Think now about a piece of magnetized accret-
ing matter, with field lines again tied to some slowly-
rotating distant point. As the matter gets close to and
crosses the event horizon, frame dragging will speed it
up, thus generating helical (Alfven) waves which move
out along the field lines. This will again generate an
outwards Poynting flux (which can mass-load and be-
come a visible jet); and the reaction force ends up mak-
ing the plasma counter-rotating as it passes through the
event horizon. Thus some of the black hole’s rotation
is lost, to supply the power carried out by the yet.
13.6.3 Duty cycles?
Accretion flows in galactic X-ray binaries provide a
possibly interesting clue. Think back to our discus-
sion of accretion disks, from last term: they are mostly
thermal (Black body) sources, possibly with an opti-
cally thin corona (could be a bremsstrahlung source).
We found that accretion disks around small (star-sized)
compact objects becomes hot enough to radiate in the
X-ray band. It turns out that X-ray emission from
galactic binaries has two states. It can be “low/hard”,
meaning lower X-ray power and a harder (nonther-
mal?) X-ray spectrum. Or, it can be “high/soft”, mean-
ing higher X-ray power and a softer (thermal) X-ray
spectrum. It turns out that steady radio jets are present
Physics 426 Notes Spring 2016 71
in the low/hard state, but not in the high/soft state.
The difference between the two states is thought to be
the accretion rate – a lower M leads to lower X-ray
power (which is constent with the simple accretion-
disk models we saw last term). Thus: perhaps a lower
M somehow changes the accretion mode in galactic
microquasars, and allows a jet to form?
It is not clear whether this scenario also applies to
AGN. There are arguments on both sides, and the ob-
servations don’t (yet?) support the existence of two
such states in AGN. If AGN do turn out to work this
way, we may have a physical origin for AGN duty cy-
cles – but to my mind, this issue isn’t settled yet.
Key points
• Jet phenomenology: what we know from observa-
tions.
• Important relativity: superluminal motion and beam-
ing.
• Basic jet “fluid physics”: collimation, propagation.
• Radio galaxy types & cartoons: FR I’s, IIs
• How does the jet affect its surroundings?
• Current ideas of jet origins – mostly MHD.
72 Physics 426 Notes Spring 2016
14 Quasars and Active Galactic Nuclei
AGN astronomy started in the 1960’s. Early radio sur-
veys had found bright, compact radio object with no
clear optical identification. Most people thought they
were simply radio-loud stars, but some thought they
were extragalctic ... in 1963 a spectrum was obtained
of the radio source 3C273. The likely optical ID was
a 13th magnitude blue object, apparently stellar (not
extended), with faint linear emission nebulosity. This
turned out to have very unusual spectra for a star, rich
with emission lines. These were thus called “quasi-
stellar objects” (QSOs), or quasars for short.
Identifying the emission lines was hard, until someone
realized the object was at the very high (at the time)
redshift, z = 0.16. After that, people started looking
seriously for these objects. The bright emission lines
and blue continuum were easy to find. By now we
know of several thousand...number counts find about
100 quasars per square degree of sky, with z <∼ 2 and
blue magnitude mB < 22 (with more, of course, at
fainter magnitudes and higher z).
14.1 Basic properties: observations
Although we now know (or think we know) that all the
varieties of AGN are driven by accretion onto a mas-
sive black hole in the core of the galaxy, that was not at
all obvious when the field started. Thus, the literature
is dominated by the unusual observational properties
of bright AGN.
14.1.1 Spectral lines
Strong optical emission lines are characteristic of (al-
most) all AGN. They are more or less the same lines
that we see in galactic nebulae – planetary nebulae and
HII regions. Emission lines from AGN seem to obey
very similar physics – and thus arise in similar con-
ditions. They are probably ionized by the observed
nuclear continuum source (the so-called “central en-
gine”; AGN are strong in the UV/soft Xray region), al-
though shock ionization is also likely in some cases.1
The striking feature of the emission lines is their width.
1Detailed analysis of the emission line strengths can tell us the
likely cause of ionization, as well as the local density and tem-
perature of the emission line gas. One piece of terminology is
necessary. Remember that quantum mechanics gives us the prob-
ability that an atom in an excited (upper) state can make a sponta-
neous transition to a lower state, emitting a photon in the process.
Permitted lines have high transition probabilities; what are called
forbidden lines have lower transition probabilities – but are not
Permitted lines (such as Hα, the n = 3−2 transition of
H) have typical linewidths ∆v = c∆ν/ν ∼ 104 km/s.
The forbidden lines (from heavy elements; OII, OIII,
NII, etc.) can be narrower, more like ∆v ∼ 103 km/s.
These widths are much too high to be due to thermal
broadening;2 these widths must be due to bulk motion
of the emitting gas clouds – either random or orbital
motion.
The difference between broad and narrow emission
lines seems to suggest that the two line types are
formed in different regions – perhaps the permitted
lines come from denser gas, closer to the central en-
gine while the forbidden lines come from less dense
gas, a bit further from the engine. Not all AGN have
this separation, however; some Seyferts have mostly
narrow-line emission, while some quasars have mostly
broad lines.
14.1.2 Continuum emission
AGN radiate in every frequency in which we’ve
looked: from radio, through infrared, to optical and
UV, thence to X-rays and γ-rays. The underlying spec-
trum is broad-band, with 2 or 3 separate features. Re-
member your radiation: telescopes generally measure
intensity, fν say, which is power/Hz. The energy in
the spectrum is better determined by∫
fνdν ∼ νfν ;
this is often what’s plotted. The total emission νfν is
typically dominated by two (or maybe 3) peaks: one
IR/optical (which itself may be two components), and
a second peak in the hard X-ray to γ-ray region. It
seems likely that the bulk of the energy comes out
in the optical/UV region, say ν ∼ 1014 − 1016 Hz.
(This dominant “bump” in the νfν plot is called the
Big Blue Bump, I kid you not.) However, some radio-
loud sources have been observed out to hard γ-rays;
their broadband spectra are dominated by the hardest
frequencies, ν ∼ 1020 − 1025 Hz (compare: what fre-
quency corresponds to the electron rest mass?)
forbidden in the QM sense. Rather, their radiative transition rates
are low enough that in dense conditions they will be de-excited by
collisions first. Thus, there is a quenching density associated with
each particular transition.2how hot would the gas be? how could ions such as OII or
OIII ever exist at such temperatures? how low would the hydro-
gen recomination rate be, and how would you ever see hydrogen
permitted lines?
Physics 426 Notes Spring 2016 73
14.2 The AGN zoo
We now know that quasars are unusual nuclei in oth-
erwise normal galaxies. Other types of nuclear activ-
ity are also possible; for historical reasons (who found
what first, in what observing band), a wide variety of
names and acronyms exist in this field. All of these
objects, as well other subclasses and acronyms I’m not
bothering to put down, are grouped together as Active
Galactic Nuclei (AGN).
14.2.1 The radio-quiet ones
Although quasars were originally detected based on
their radio emission, radio-loud quasars turn out to
be rare, something like 10% of the quasar population.
Radio-quiet quasars (usually “QSO’s”) are domiated
by their bright core, with strong optical emission lines
as well as strong continuous emission (optical/UV, also
X-rays).
Seyfert galaxies and Markarian galaxies3 are spi-
ral galaxies with nuclei which are quasar-like in their
broadband and line emission properties, but not so
bright as the quasars. Seyferts are never radio-loud;
they have weak radio cores and can have small, weak,
poorly collimated radio jets which do not propagate out
of the galaxy (due to the denser ISM? due to different
conditions in the core?) Weaker versions of Seyferts,
also in spiral galaxies, are called LINERS (Low Ion-
ization Nuclear Emission Region) – with emission
lines but less nonthermal continuum than the Seyferts.
14.2.2 The radio-loud ones
As we saw in the previous chapter, The term “radio
galaxy” refers to an elliptical galaxy with a radio jet
and double-lobed (or tailed) radio structure on super-
galactic scales (linear extent, side-to-side, can range
from ∼ 100 kpc to a few Mpc). The term “radio-loud
quasar” (QSR) refers to a quasar with the same sort
of radio jet-lobe structure. The distinction is mainly a
question of how strong the nuclear activity is (as seen
by us, anyway): how strong is the nonthermal con-
tinum and/or the emission lines?
3both named for the person who originally cataloged them –
based on unusually strong nuclear emission lines or UV contin-
uum.
14.2.3 Blazars and friends
These are a subset of the radio-loud ones (about 10%)
with unusually bright, active, variable radio-to-optical
cores. People in this field speak in acronyms:
OVVs (optically violent variables). These are the
highly variable, clearly relativistic ones. They show
a flat radio spectrum (suggestive of a compact, syn-
chrotron self-absorbed core and a pc-scale radio jet
with a few bright knows). They are highly variable,
timescales from days to years.4 These are the ones that
show clear superluminal motion.
BLLs, as described, are named for BL Lacertae (an
object previously thought to be variable star locally).
The emission lines are quite faint compared to the
cointinuum. As with OVVs, the BLL radio emission
is strongly polarized, flat spectrum, & highly variable
(δt >∼ days).
Both of these are often grouped together as blazars.
The general picture – motivated by the observations of
superluminal motion – is that we are looking “down
the pipe” of the jet, close to its axis. In addition to ex-
plaining the superluminal motion, such a geometry will
enhance the variability (by relativistic effects and by
the Doppler-beaming effects on a jet with some slight
inhomogeneities.
14.2.4 Parent galaxies
What are the parent galaxies? We noted above that
radio galaxies are found in elliptical galaxies, while
Seyferts (and related radio-weak AGN) are found in
spirals. What about quasars? There has been a sense
that radio-loud quasars live in E’s, while radio-quiet
live in S’s. This view seems to be changing, however,
at least for the brightest of the population. Quoting
from Dunlop etal (2003): “Virtually all powerful AGN
live in normal, massive ellipticals: the parents agree
with local bright E’s in morphology, luminosity, scale-
length, consistency with fundamental plane, axial-ratio
distribution, and colors (evolved stellar population, age
10-13 Gyr).” “The inevitable conclusions is that these
galaxies, or at least most of their stars, must have
formed at high redshift.” “Spheroidal hosts become
more prevalent with increasing nuclear luminosity; for
bright enough nuclei, the hosts of both radio-loud and
4There is an argument currently going on about variability on
much shorter δt’s, less than a day; these are the Intra-Day Vari-
ables. The variability is probably due to interstellar scintillation,
although some people want it to be intrinsic.
74 Physics 426 Notes Spring 2016
radio-quiet AGN are massive ellipticals.”
14.3 The usual model: a massive BH
We know the answer of course (or think we do): all of
this comes from accretion onto a massive black hole.
Let’s see how (or if) this model can explain the variety
of observations.
14.3.1 Zoom in: the central kpc and within
The first clues may have come from the emisison lines.
In most objects we need photoionization; with detailed
modelling one can pinpoint the photon density, which
means the distance from the central engine. Applying
this to the broad and narrow lines shows that the NLR
must be at ∼ kpc from the engine, while the BLR is
closer, more like pc to tens of pc away. Both broad and
narrow line emission must be “patchy”, for instance
coming from dense clouds.5 It may not be correct,
however, to picture the clouds as uniformly distributed
over the entire central kpc. We are learning, from HST
imaging, that the emission-line gas in the central kpc is
often confined to emission “cones” – what you would
expect, for instance, if the ionizing radiation escaped
from the central engine only in a moderately narrow
cone.
14.3.2 Zoom in further: the central pc and within
This is, of course, the region of the central engine; the
region we probe indirectly through variabilty, and di-
rectly through milliarcsecond radio imaging (as with
the VLBA). The generic picture is, as we’ve discussed,
accretion onto a massive black hole, probably with a
jet being driven out the rotation axis. Figure 14.1 il-
lustrates the general thinking. The accretion disk is
mostly thermal, that is resembling the models you saw
in P425: spatially thin and optically thick. It emits
as a black body, and is thought to be the origin of the
“big blue bump”. Nonthermal emission – high energy,
X- and γ-rays – comes from an optically thin region
somewhere close to the black hole and the inner region
of the disk; very high gas temperatures, possibly rela-
tivistic plasmas, and electron-positron physics may be
going on here. A two-sided jet is driven out; if this is in
5Why? One, we can see the engine, so the line-emitting gas
doesn’t completely cover it. Two, we know the density of the line
emitting gas, from quenching and line-ratio arguments; comparing
this to the total volume at the clouds’ distance also requires the
volume be incompletely filled.
an elliptical the jet propagates to extragalactic scales.
The gas emitting the broad emission lines is sketched
as discrete clumps, moving at high speed (here guessed
to be random). The intercloud region is probably a hot
wind, driven out from the accretion disk by a combi-
nation of its own temperature and magnetic/centrifugal
effects.
jet
jet
thermal emission
(big blue bump)
massive
black hole
nonthermal emission
(thermal) accretion disk
emission line clouds
Figure 14.1 Generic cartoon of the central pc of the AGN.
An accretion disk feeds a massive black hole. Details are in
the text. The scale of this cartoon is a few pc (remember the
event horizon of the black hole is only ∼ 1− 10 AU.
14.3.3 Why radio-loud vs. radio-quiet?
We still don’t have much of an answer to this ques-
tion, or even a standard “toy model”. Radio galaxies,
and maybe all of the brightest quasars, live in ellipti-
cals; Seyfert nuclei, and maybe most of the radio-quiet
quasars, live in spirals. Why? I wish I knew.
14.3.4 Why are only some galaxies “active”?
This one is even harder to answer. We now know that
every galaxy contains a massive black hole at its core
(check back to chapter 1 for discussion). But this black
hole is “active” in only a few per cent of galaxies.
Why? Once again, I wish I knew.
14.4 Unification Models
Can these various disparate categories and acronyms
be explained by one simple, unified picture? Maybe,
maybe not .. this issue has strong adherents, because
simple pictures are pleasing and attractive. But one
can push too much for simplicity, at the expense of ig-
noring some of the physics ... some of the community
belong to “the unification church”, while others (in-
cluding your author) remain agnostic.
There seem to be two important motivations for this
view, as follows.
Physics 426 Notes Spring 2016 75
14.4.1 Relativistic beaming
We’ve already met this several times. If the jet is driven
out at relativistic speeds, with Lorentz factor γ – and
we know it is, from the observation of superluminal
motion – then a viewing angle within an angle ∼ 1/γof the jet axis is special. From this vantage, only, we
will see vapp > c, strong forward beaming, enhanced
variability, and etc – that is we will see a blazar.
14.4.2 Obscuration and tori
We noted above that the optical line emission is
anisotropic in some nearby AGN, being located in an
ionization cone. Such a cone might arise from a fat
torus (think of a donut) surrounding the central engine,
allowing hard photons to escape only in a fairly broad
cone along the axis of the torus. In addition, emis-
sion lines from Seyferts (which are nearby, and bright,
enough to do this measurement) appear different in po-
larized than total light. Some Seyferts that show only a
narrow line region (it’s called a Seyfert 2, if you care)
turn out to have broad line wings when seen in polar-
ized light. Remember that Thompson scattering polar-
izes the light. Thus, if the light from the central en-
gine, which sits at the heart of the donut, is scattered
(by some diffuse ionized gas, above or below the sys-
tem), an observer can detect the broad lines – which
come from very close to the black hole – at angles at
which the direct emission is obscured by the torus.
side view: see only the torusside view: see only
the torusside view: see only
tne obscured engine
top view: see the engine
top view: see the engine
the "dusty torus"
path of scattered (polarized) light, revealing inner engine
Figure 14.2 Extremely generic cartoon of the unified
model. The central engine may be surrounded by a fat,
opaque torus (maybe containing dust). Whether or not you
see the central engine depends on your viewing angle. The
scale of this cartoon is a few hundred pc.
14.5 AGN demographics
To summarize: we’ve walked through the general pic-
ture of an AGN: a massive black hole sits in the core
of a galaxy. Galactic material (ISM) accretes onto the
black hole. Side effects of that accretion are the gener-
ation of a strong radiation over all frequency bands (by
a mix of thermal and nonthermal processes), and the
ejection of collimated plasma jets (which may or may
not propagate out of the galactic core).
The remaining questions, then, are demographic. How
common are AGN, and how were they distributed over
the history of the universe? Why does every bulge con-
tain a black hole, and which formed first (the bulge
or the BH)? Does every nuclear black hole make an
AGN? How did the black hole come to exist in the
galactic core (is it the chicken or the egg?) How have
these AGN evolved over the course of the universe? To
answer these questions, we need to go to high redshift
- z ∼ 2− 10, say, to see what happened at early times.
These questions are far from being answered, but we
do have some important clues. In case you’re not fa-
miliar with high-redshift thinking, I’ve put some basic
cosmography in an Appendix to this chapter.
14.5.1 Was there a “quasar era?”
Quasars – like other AGN – are rare in today’s uni-
verse; most galaxies do not contain a currently active
central engine (even if, as we now know, they proba-
bly contain a massive BH). This is not the case at early
epochs, however. Two facts become clear from quasar
surveys.
• At a given epoch – today, for instance – there are
more faint quasars than bright ones. The number
per volume at an absolute (optical, blue) magnitude
MB can be written very approximately as Φ(MB) ∼ΦoM
−0.7B .
• The absolute number of quasars – either at a given
magnitude, or integrated over all luminosities – was
greater at early epochs. (This is the same thing as say-
ing that the constant above, Φo, is a function of z.) The
space density of quasars rises out to z ∼ 2, has a broad
peak in the range z ∼ 2−3, then decays again at higher
z. This epoch, which saw a lot of quasar activity, is
called the “quasar era”.
What has changed, with quasars, since the quasar era?
What is this evolution of the number density saying?
Up to a few years ago, there were (at least) two alter-
native possibilities. One is that the fraction of quasars,
per galaxy population, has stayed the same with time,
but the luminosity of a given quasar was much brighter
at z ∼ 2 than it is now. Because all quasar surveys
must be flux limited, we would be detecting more of
76 Physics 426 Notes Spring 2016
them early on. This is called luminosity evolution. An
alternative theory is that the mean quasar luminosity
has not evolved with time, but that there were simply
more of them back then ... so that the increase in mea-
sured numbers of quasars back to z ∼ 2 is what it
appears to be. The real answer is probably somewhere
inbetween these two extremes...but we must remem-
ber that almost every nearby galaxy (at least) harbors a
black hole. If we define this as a dark quasar, then lu-
minosity evolution must be the preferred model. How-
ever, we are learning more about galaxy formation and
evolution, and the question may be more complex.
14.5.2 What about galaxy formation?
The quasar counts, described above, have been known
for some time. A currently very active research area
tries to understand when (at what z) galaxies formed.
This is much harder than counting quasars (which are
easy to pick out, they are bright and small and blue).
How can we pick out galaxies in the act of formation?
Think about a protogalaxy: picture a self-gravitating
clump of stuff collapsing due to its own gravity, and/or
being enhanced by the accumulation of smaller-sized
clumps (mergers). The baryonic matter will lose en-
ergy (by radiation), fall through the dark matter, and
must eventually form stars. We know ellipticals have
little active star formation today; their stars must have
formed in one great rush, at early times. Spirals today
do have ongoing star formation, but they probably also
had a strong starburst phase when they first formed.
So, the general idea is to look for unique signs pointing
to early bursts of star formation. We know a lot about
local star formation regions. Optically, they are bright
in Hα, and also in the UV (due to all those hot young
stars). Moreover, we know they are dust-enshrouded,
and thus very bright in the IR or sub-mm band. Thus:
we can try to find distant objects that are bright in the
UV, or in Hα, or – the currently most promising – in
the IR/sub-mm.
The tentative result of these surveys (the work is still
ongoing, and the arguments as large as the error bars at
high z) is very interesting. Given the variety of obser-
vational probes, which mix apples and oranges, people
generally turn their data into “the rate of star formation
as a function of z”.6 This is called a “Madau plot”,
6This can be a long daisy chain. For instance: measure an IR
luminosity, say; from that determine how many bright, massive
stars are needed to power the dust; divide by the main sequence
after the person who first did one.
Difficulties in the analysis aside, everyone in this field
agrees on the low-z result. The number of strongly-
starbursting galaxies increases with redshift, out to z ∼2 – which is just the quasar era. The uncertainties come
at higher redshift: it is not clear if the star formation
rate declines again, at higher z (like the quasars), or
continues more or less steady (out to, say, z ∼ 5, the
current limit of this sort of work).
14.6 Ending with questions
All of this raises some intriguing questions, which
seem as good a way as any to end these notes.
• Did the bulk of galaxy formation take place at the
same time as quasars were most active?
• Is a strong starburst a part of the mass-accretion pro-
cess which kept the AGN powerful at that epoch?
• Are the merger and dissipation events that made the
bulge or E galaxy the same events that made the quasar
shine?
• Are the majority of nuclear BH dark today because
they aren’t being fed? Are galaxy mergers necessary to
transport the gas (i.e. BH food) close to the nucleus?
And ... that’s it, folks! It’s been fun – have a good
summer!
Key points
• The phenomenology: important observational
trends..
• The usual model: what’s there on sub-pc scales?
• “Practical cosmology”: what does “high z” mean
(ages, timescales, etc)?
• The QSO epoch and how it might connect to galaxy
formation.
lifetime of the massive stars; and you have “derived” a “star for-
mation rate”. Or ... measure the radio luminosity; use the fact
that the radio and IR powers are well correlated for nearby spiral
galaxies, to estimate the IR luminosity; and return to top..
Physics 426 Notes Spring 2016 77
14.7 Appendix: a little practical cosmology
Before we can talk about BH and AGN in the early
universe we need to review the language used. The
context is cosmology: we work with the scale factor,
R(t). This is a quantity which describes (“scales”)
the distance between two fixed points in an expand-
ing universe – say two nearby galaxies. The evolution
of R(t) reflects the fight between gravity (an attrac-
tive force), initial conditions (how much expansion did
the universe start with?), and the effect of any vacuum
energy density (the cosmological constant; “dark en-
ergy”). For our purposes here, let’s assume that R(t)is a known function (found from the solution of Ein-
stein’s field equations). Everything we can measure
about a distant object – size, luminosity – depend on
how much the universe has expanded between the time
it emitted a light signal (tem) and the time we receive
that signal (today; usually called to).
14.7.1 Just what is the redshift?
The redshift is defined is 1 + z = δλ/λ (the shift in
an emission wavelength, say of a spectral line, relative
to its rest value). The redshift of a cosmological object
is an important quantity. It tells us not just the reces-
sion speed (which isn’t very interesting), but – more
importantly – the distance and age of the object.
There are at least 3 physical causes of a redshift.
• You remember the simple Dopper shift: for low
speeds, 1 + z = v/c; and for relativistic speeds, the
Lorentz transform in simple geometry gives 1 + z =γ(1 + β).
• There is also a gravitational redshift, which occurs
when light emitted in a potential well (say at the sur-
face of a star) climbs out to the observer. For a
Schwarzchild geometry this becomes 1 + z = 1/(1 −rs/r)
1/2.
• Finally, there is a cosmological redshift: the fre-
quency of a signal decreases due to the expansion of
the universe while that signal propagates to the ob-
server. This is the one we want here: it becomes
1 + z = R(to)/R(tem).
Because the cosmological redshift is intimately tied to
the expansion of the universe, it becomes a handy way
to describe the distance and age of an object. Two
things are worth noting here.
14.7.2 The Hubble diagram
Think about a source at some distance D. In Euclidean
space its light spreads out over a surface of area 4πD2
before it gets to the observer, so that one sees a flux
∝ L/D2. But, this changes if space is not Euclidean.
The key fact is that the area of the D-sphere is not
4πD2. The details of how the area changes depend on
the cosmological parameters (curvature and cosmolog-
ical constant), but the general trend can be sketched, as
in Figure 15.1. NOTE that for z ≪ 1, curvature effects
don’t matter; dl ∝ z, just as in Euclidean space. The
original Hubble’s law (v = Hoz) applies in this limit.
Also note that the so-called Hubble constant, Ho, is the
slope of this curve as z → 0 (that is, it isn’t a constant,
but a variable).
zz
lum
inos
ity d
ista
nce
mag
nitu
de
0 1 5 0 1 5
Figure 14.3 How the curvature of space affects the lumi-
nosity distance, dl (defined so that the observed luminosity
of a source at redshift z is L/4πdl(z)2). Note that for small
redshifts, dl ∝ z, and the classical Hubble’s law is recov-
ered. The right hand diagram is often called the Hubble
diagram; its high-z extension can, in principle, be used to
measure cosmological parameters.
z
age
/ H
ubbl
e ag
e
look
back
tim
e / H
ubbl
e ag
e
z
Figure 14.4 How the age of the universe, and of an object
we observe, depend on redshift. Left: the age of the universe
(counted from t = 0, the big bang) when an object had red-
shift z. Right, the lookback time – the difference between
the age of the universe at z and the age of the universe now.
This tells us how long ago an object existed, if we see it at ztoday. The “Hubble age” is defined as 1/Ho.
14.7.3 The lookback time
How long ago did an object exist, if we see it now at
redshift z? If space were Euclidean, that would be
simple: (age at emission) = (today’s age) - (distance
78 Physics 426 Notes Spring 2016
/ lightspeed); and the lookback time would be propor-
tional to the redshift. But as with the Hubble diagram,
space curvature makes this more interesting for z >∼ 1.
Numbers: z = 2 gives a lookback time ∼ 0.5/Ho;
z = 5 gives a time ∼ 0.6/Ho, and for higher z’s the
lookback time doesn’t change by a lot.